Artificial Intelligence Innovations in Cerebrovascular Neurosurgery: A Systematic Review of Cutting-edge Applications

doi:10.21203/rs.3.rs-4435188/v1

Download PDF

Research Article

Artificial Intelligence Innovations in Cerebrovascular Neurosurgery: A Systematic Review of Cutting-edge Applications

https://doi.org/10.21203/rs.3.rs-4435188/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Introduction: Artificial Intelligence (AI) offers transformative potential for cerebrovascular neurosurgery, enabling novelapproaches to predict clinical outcomes, enhance diagnostic accuracy, and optimize surgical procedures. Thissystematic review explores AI's integration and impact in this specialized field, emphasizing improved patient careand addressing challenges in resource-limited settings.

Methods: A comprehensive literature search across PubMed, Scopus, and Web of Science gathered peer-reviewed articlesdetailing AI applications in cerebrovascular neurosurgery. Studies were analyzed to evaluate AI effectiveness inpredicting outcomes, enhancing detection/diagnosis, facilitating risk stratification, and integrating into surgicalworkflows. Ethical implications and applicability in resource-limited settings were examined.

Results: AI demonstrates significant potential in predicting clinical outcomes by analyzing large datasets to forecast patienttrajectories. In detection and diagnosis, AI algorithms utilizing machine learning show high accuracy in interpretingneuroimaging data, enabling earlier and more precise cerebrovascular condition diagnoses. Risk stratification isenhanced through AI's ability to classify patients based on risk profiles, enabling better resource allocation andtargeted interventions. However, AI integration into surgical workflows requires substantial adjustments.

Conclusion: AI in cerebrovascular neurosurgery presents a promising avenue for advancing patient care through personalizedtreatment strategies and improved diagnostic and predictive accuracy. Ethical considerations regarding data privacy,algorithmic fairness, and equitable distribution must be rigorously addressed. Future research should focus onovercoming challenges, enhancing clinical workflow integration, and ensuring accessibility across diversehealthcare settings. Continuous collaboration between engineers, clinicians, and ethicists is advocated to fosterinnovative and ethical AI applications in neurosurgery.

Cerebrovascular diseases like ischemic and hemorrhagic strokes, aneurysms, and arteriovenous malformations carry a substantial global health burden. Prompt identification, risk evaluation, and development of treatment plans are essential for enhancing outcomes and limiting disability [1]. However, the complexities involved pose major challenges, especially in resource-limited settings. In this context, artificial intelligence (AI) stands as an innovative technology capable of transforming cerebrovascular care by improving diagnostic precision, optimizing clinical choices, and enhancing patient outcomes [2].

In the realm of cerebrovascular surgery, AI techniques like machine learning and deep learning have found diverse clinical applications. These include automated detection of stroke and cerebrovascular lesions on neuroimaging, precise segmentation of pathologies like aneurysms, prediction of outcomes after interventions, and guidance on optimal treatment strategies [3]. By utilizing AI's potential to analyze vast and intricate datasets and identify complex patterns, it can complement human expertise in areas like risk stratification, surgical planning, and clinical prognostication [4]. Overcoming limitations in human cognitive capacity, AI promises to reduce diagnostic errors, expedite care pathways, and enable more personalized cerebrovascular disease management [5].

Despite AI's promising potential in cerebrovascular surgery, implementation faces challenges like data limitations, regulatory barriers, and the need for robust validation studies. Concerns over interpretability, transparency, and seamless clinical integration must also be addressed [6]. This systematic review comprehensively analyzes the literature on AI applications in cerebrovascular surgery, highlighting strengths, limitations, and future directions. By synthesizing evidence, we aim to inform clinicians, researchers, and policymakers about this rapidly evolving field, identifying opportunities to enhance cerebrovascular disease care and improve outcomes.

Research Question and Selection Criteria

The present systematic review was conducted based on the research question “What is the efficacy of artificial intelligence techniques in cerebrovascular neurosurgery?” The systematic search and identification of eligible studies were centered on the PICO criteria elaborated in Table 1. The present systematic review was registered with the International Prospective Register of Systematic Reviews (PROSPERO), with reference ID: CRD42023454067

Databases included and Search Strategy

The search was conducted in three databases: PubMed, Scopus and Web of Science on 7 June 2023. We also undertook a search in the pre-print servers such as medRxiv, arXiv, bioRxiv, BioRN, ChiRxiv, ChiRN, and SSRN. In addition, studies obtained by hand search in the references of the eligible primary research papers and reviews, which satisfied our eligibility criteria, were also included in the data extraction. The search keywords included “artificial intelligence”, “cerebrovascular neurosurgery”, “cerebral aneurysms”, “arteriovenous malformations”, “ischemic stroke”, “subarachnoid haemorrhage”, and related terms. The database-wise search strategy applied and the results obtained have been enumerated in Table 2. Mendeley Desktop V1.19.5 software was utilized for importing the articles, managing the citations, removing duplicates, and coordinating the overall review process between the authors (Table 2).

Screening of studies

Title and Abstract screening

Two authors (PR & TB) independently reviewed the title abstracts of the studies obtained from the above systematic search by applying the eligibility criteria and identified articles for full-text screening. If there was a disagreement regarding the inclusion of a study for full-text review, the co-authors conversed among themselves to build consensus and decided on the eligibility.

Full-text Screening & Data Extraction

Eligible full-text articles were reviewed for suitability of data extraction by two authors, and data extraction was done by them, independently (PR & TB). Contradictions in data extraction between the authors were removed in a consensus meeting conducted at the end of the independent extraction. Irresolvable contradictions were adjudicated by the third author (SG). A final table was formulated that included information such as the name of the first author, publication year, the geography of the study where it was conducted, the design of the study, the total number of patients, and the Preferred Reporting Standard of Systematic Reviews and Meta-Analysis (PRISMA) flowchart and checklist were utilized to report the overall process of search, screening, data extraction, systematic review, and the meta-analysis conducted to ensure scientific precision (Figure 1).

Quality Assessment

The quality assessment of the included study was independently done by two authors (PR & TB) using the ROBINS-I tool for non-randomized studies. The Author (SG) solved any disagreement in their assessment. The overall risk of bias in individual studies was categorized based on the criteria outlined in the methodology section.

Summary measures and synthesis of results

It was predetermined to use Review Manager (RevMan, version 5.3, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014) to pool the data of an outcome with a high degree of similarity for quantitative analysis. The mean difference and 95% confidence interval (CI) were used for the continuous data. The statistical heterogeneity was intended to be tested using Tau² and I² statistics. It was predetermined that the random-effect model would be used if the I² was more than 50%. It was planned to use a detailed narrative summary of the findings if there was considerable clinical and methodological heterogeneity across the included studies.

ACUTE ISCHEMIC STROKE

There are 20 studies which deals with application of Artificial Intelligence and machine learning models in modalities like detection, outcome prediction and risk prediction associated with Acute Ischemic Stroke [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], Table 3.

Outcome prediction

Yu Bai et al in their study compared the long-term clinical outcome in VR-based limb rehabilitation versus traditional drug therapy in stroke patients, the study found a beneficial effect on AI-based technology [9]. Jacob R Morey et al. in another study studied the effects of Viz application in decreasing the time to treatment in LVO AIS patients, they found that a significant reduction in time was for response initiation accompanied by fewer variations[10]. The mean door-to-skin puncture time was also reduced following post-viz implementation, although the results were not statistically significant. Dougho Park et al. in their study used different machine learning models for screening the risk of aspiration in patients hospitalized with acute stroke which yielded beneficial and practical results [11]. Using similar models Kai Wang et al developed a similar model for the prediction of stroke recurrence within one year in AIS patients, the study implied that the Random forest (RF) model is the most accurate one to predict recurrence [14]. Takeshi Imura et al. in their study assessed the likelihood of home discharge following inpatient rehabilitation by taking into account numerous functional and environmental factors using CART models, observed that for identifying home discharge, the CART model that included fundamental data, functional parameters, and environmental factors had the highest accuracy [22]. The analysis found the first single discriminator for home discharge as FIM dressing of the upper body. Martin Bretzner et al. used the ElasticNet model to extract information regarding relative brain age and comparing with chronological age to predict the functional outcomes of stroke [23]. The study observed greater relative brain age (RBA), a marker of older-appearing brains, was linked to an enhancement of cardiovascular risk factors and less favourable functional outcomes following stroke. Jinhao Lyu et al. studied collateral-core ratio (CCR) to predict clinical outcomes (XGBoost) using patient features, assessing the significance of variables, and interpreting model predictions (SHAP), found that compared to the model without CCR, CCR had a greater discriminatory ability in predicting unfavorable outcomes [26]. Lohit Velagapudi et al. (6) utilized the scikit-learn package in Python to investigate the efficacy of supervised machine-learning algorithms in predicting first-pass reperfusion and conducting feature analysis in patients undergoing mechanical thrombectomy [12]. The study revealed that the Random Forest and Support Vector Machine (SVM) models achieved accuracies of 67.1%, outperforming the logistic regression model, which attained an accuracy of 65.8%. According to the Random Forest model, the top five predictors of first-pass reperfusion were aspiration, hyperlipidemia, hypertension, utilization of a stent retriever, and the duration between symptom onset and catheterization.

Detection

Yahav-Dovrat in their study implemented the Viz LVO algorithm to detect occlusions from the ICA terminus to Sylvian fissure but the system showed suboptimal sensitivity, which prevents it from being used as a diagnostic tool [8].

Tatsat R Patel et al. in their study used multivariate machine learning models to predict successful outcomes for ADAPT in middle cerebral artery (MCA) stroke using pre-treatment imaging metrics and found the logistic regression model achieved the highest accuracy (74.2%) in the testing cohort [15]. Michelle Livne et al. studied high-performance Vessel Segmentation Using Deep Learning Model U-Net in Patients With Cerebrovascular Disease [16]. The study found that both the complete and the reduced architecture produced good performance from the U-net versions also, the visual study showed that large vessels performed exceptionally well and tiny vessels adequately. In a small number of cases, pathologies such as cortical laminar necrosis and a rete mirabile caused poor segmentation performance. Jia You et al. studied the ability of the suggested Dissimilar-Siamese-U-Net (DSU-Net) model to autonomously segment the hyperdense middle cerebral artery sign (HMCAS) from non-contrast computed tomography (NCCT) scans of the head [19]. Results show that the proposed DSU-Net outperforms the baseline U-Net and numerous state-of-the-art models for clinical practice and offers a fresh strategy for HMCAS automatic segmentation. JunHua Liao et al. compared the performance in aneurysm detection of the developed ML model with that of 3 existing frameworks, findings suggest the performance of the frameworks can be enhanced by adding more geographical and temporal information [21]. As a result, when compared to the other frameworks, the bi-input+RetinaNet+C-LSTM framework performed the best. Our research shows how our method can help doctors identify cerebral aneurysms on 2D DSA pictures. Stavros Matsoukas et al assessed the precision of AI software in a three-tiered multihospital stroke network in real life [12]. The study found that accuracy metrics support Viz LVO as a useful adjunct tool in stroke diagnostics.

Risk prediction

Hidehisa Nishi et al. used the gradient boosting tree algorithm to predict the risk of cerebral infarction in patients with atrial fibrillation and reported that their machine learning model had a higher predictive accuracy than the CHADS2 and CHA2DS2-VASc risk scores [7]. Sunil A Sheth et al. used DeepSymNet to evaluate large vessel occlusion (LVO) and ischemic core volume in patients using a widely available modality, computed tomography angiogram (CTA) [13]. These findings show that CTA may contain the data necessary to perform neuroimaging evaluation for endovascular therapy with accuracy comparable to that of sophisticated imaging modalities, and that machine learning can automate the analysis. Hidehisa Nishi et al. compared the technique with previously created pretreatment scoring systems to predict the clinical outcome of large vessel occlusion (LVO) before endovascular therapy. They reported that the clinical outcomes of patients with anterior circulation LVO undergoing mechanical thrombectomy can be predicted more accurately by machine learning approaches leveraging numerous pretreatment clinical characteristics, as opposed to previously developed pretreatment scoring techniques [7]. Yoonjae Chung et al. studied the utilization of optical diffraction tomography (ODT) and deep learning (DL), automated the process of histologically quantifying the composition of thrombi reported that this method will make it possible to quickly and quantitatively assess the makeup of blood clots and, speeding up preclinical research and the detection of cardiovascular disorders [18]. Kai Wang et al. explored the predictors of death in acute ischemic stroke (AIS) patients within 1 year based on machine learning (ML) algorithms and found that clinicians may use the network calculator established in this study, which is based on the XGB model, to assist them make more individualized and logical treatment decisions [20]. Pum-Jun Kim et al. performed a binary logistic regression (BLR) and association rule mining (ARM) to analyze the data and investigate the relationship between obesity and stroke outcomes found that at three months, the outcomes of patients with acute ischemic stroke (AIS) are improved by obesity [25]. Obesity had a favorable effect on the result, according to the binary logistic regression (BLR) study. In addition, the association rule mining (ARM) study showed that patients with obesity who had positive outcomes also had mild stroke severity and a younger age.

INTRACRANIAL ANEURYSMS

After screening, we identified 21 studies related to the management of Intracranial Aneurysm (ICA) [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46]. They consisted of eleven studies on diagnosing ICA [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], one study on treatment outcomes [38], nine studies on risk assessment [39], [40], [41], [42], [43], [44], [45], [47], and one study on prediction of post procedural complications [46], Table 4.

Diagnosis of ICAs and Vessel Segmentation

Ou C et al. developed a cross-scale dual-path transformer module for diagnosing ICA in patients that achieved 100% accuracy in parent artery classification and accurate morphology measurement [27]. In another study by Ou C et al., a PyRadiomics module used in ICA patients indicated a difference in radiomic signatures between ruptured and unruptured aneurysms [28]. Wang J et al. and Liu X et al. both used a 3D convolutional neural network (CNN) model which achieved a high sensitivity in automated detection of ICAs, with the later study demonstrating a positive predictive value for detection and segmentation of medium and large sized ICAs [29], [30]. Duan H et al. proposed a 2 stage CNN architecture to detect ICA on 2D-DSA images which was faster and more accurate than existing models [31]. Jin H et al. developed a fully automated detection and segmentation deep neural network-based framework, for evaluation and contouring ICA from 2D+time digital subtraction angiography (DSA) sequences which successfully detected 89.3% ICAs with a patient level sensitivity of 97.7% [32]. Benvenuti L et al., used volume rendered helical computed tomographic (CT) angiographic data sets by use of a surgical navigation technology reported a time- and cost-effective method of performing automated 3D volume rendering useful in emergency cases. The model detected precise location and anatomic features of the ICA [33]. You W et al. developed a 3D-Unet Algorithm for aneurysm detection and segmentation. The preliminary results of this study indicate a great potential for the use of AI in treatment of ICAs [34]. Poppenberg KE et al. constructed prediction models using Least Absolute Shrinkage and Selection Operator (LASSO) selected transcripts from Neutrophil RNA sequencing. Models trained using these transcripts had maximum accuracy of 90% in predicting the presence of unruptured ICAs [35]. Su J et al. developed a machine learning model based on a metabolic biomarker panel consisting of lactate, glutamine, homoarginine, and 3-methylglutaconic acid levels which displayed satisfactory diagnostic and risk assessment availability [36]. Veeturi SS et al studied the relationship between hemodynamics and pathobiology of ICA using machine learning which indicated that elevated wall shear stress was associated with thin regions of ICA wall [37].

Treatment of ICAs

Liu C. et al. developed an AI algorithm to automatically generate 3D microcatheters using data from 3D rotational angiography and digital imaging and communications in medicine (DICOM). Thirty patients successfully completed the procedure without perioperative complications with a satisfactory stability of the microcatheter [38].

Risk Assessment

Heo J.et al. developed a model using several algorithms and compared them. The XGB model displayed best AI risk prediction in patients at risk of developing ICAs [39]. In a similar study conducted by Ou C et al., the XGB model displayed superior results compared to LR model and a PHASES score method [47]. In another study, Ou C et al. tested the feasibility of the automated machine learning (AutoML) to develop high-quality ML models for risk prediction. The AutoML derived ML model displayed better performance compared to statistical and manually trained models [40]. Li P et al. proposed an end-to-end deep learning model, a multiscale 3D CNN that resulted in 10-15% improvement in accuracy of rupture status prediction [41]. Liu J et al. used a two-layer feed-forward artificial neural network (ANN) which resulted in an overall prediction accuracy of 94.8% in anterior communicating artery (ACOM) aneurysms [42]. Shiraz Bhurwani MM et al. used Deep Neural Networks (DNNs) and Angiographic Parametric Imaging to predict ICA occlusion by resulting in an average prediction accuracy of 62.5%,70.8%, and 77.9% for the un-normalized, normalized, and relative subgroups. It demonstrated the feasibility of using this model in pre- and post-operative imaging [43]. Tong X et al. used machine learning-based cluster analysis for discriminating the risk of rupture of small multiple ICAs. The algorithm was able to identify three distinct clusters with variable subarachnoid hemorrhage risks [44]. Lu T et al. used a weighted gene co-expression network analysis (WGCNA) and machine learning integrative approach to conduct risk assessment. A four-gene machine learning-derived gene signature (MLDGS) was used. This signature identified patients with high risk of aneurysm rupture [45].

Prediction of Periprocedural Complications

Tian Z et al. developed a prediction model using machine learning to investigate periprocedural complications using artificial neural network, random forest (RF), and logistic regression (LR) algorithms [46]. ANN and RF models deliver satisfactory performance, indicating that ML is a valuable tool for prediction of perioperative complications after endovascular treatment for unruptured ICAs.

Hemorrhage

Nawabi et al. (1) used machine learning algorithms to examine non-enhanced CT scans of individuals with acute intracerebral haemorrhage (ICH) [48]. Comparable to multidimensional clinical grading systems, machine learning-based evaluation of quantitative high-end picture features was able to predict functional outcome. A statistically significant improvement in AUC resulted from the combination of traditional scores with picture characteristics. Tang et. al. similarly used the ML model to predict hematoma expansion in intracerebral hemorrhage (ICH) patients , the study concluded that Deep learning technology in this study can accurately classify more than 90% of hematomas with or without enlargement, which is superior to existing approaches based solely on clinical factors [49]. From non-contrast CT scan images, deep learning technologies may be able to forecast hematoma expansion favourably. Park et al. examined the potential use of near-infrared spectroscopy (NIRS) to detect delayed cerebral ischemia (DCI) in patients with subarachnoid haemorrhage (SAH) [50]. The study found that although changes in NIRS were not statistically significant, they tended to produce greater diagnostic accuracy than transcranial Doppler ultrasonography (TCD). Real-time DCI detection can be accomplished with NIRS. Duan et al. suggested a CNN-based architecture to automatically identify cerebral aneurysms from 2D-DSA images [31]. The outcomes showed that, in comparison to the previous research studies of traditional DIP approaches, our proposed two-stage CNN-based architecture was both more accurate and quicker. Overall, our study provides evidence that utilising CNN, it is possible to help medical professionals find cerebral aneurysms on DSA pictures. Rangaraj et.al. used a single pipeline of a multi-task model is proposed for the segmentation and risk assessment of haemorrhages from beginning to end [51]. Hu et. al. used LR and ML models and assess their ability to predict delayed cerebral ischemia (DCI) following aneurysmal subarachnoid haemorrhage (aSAH) [52]. This multicenter investigation discovered that a number of ML techniques, in particular RF, outperformed traditional LR. In addition, a web-based predictor tool based on the RF model was created to locate patients at high risk for DCI following SAH and enable prompt therapies.

AV malformation

Hong et al. in their study YOLOv5 and YOLOv8 algorithms for detecting brain arteriovenous malformations in U-Net and U-Net++ models and differentiating its nidus in MR Angiography [53]. The outcomes demonstrated that pretraining and augmentation significantly improved the outcomes in YOLO model.

Thrombosis

Yang et. al. used a deep learning (DL) model algorithm for detection of cerebrovascular thrombosis in the routine MRI done as per standard protocol [52]. The study was done among 392 participants which included 294 patients with CVT and 98 without CVT. These patients were divided into two groups: the test set consisting of human intervention and the development set using ML. With high sensitivity and specificity, the CVT-detected DL algorithm shown here improved diagnostic performance of standard brain magnetic resonance imaging, offering a potential method for identifying CVT. While Nishi et al.(2) analyzed the effectiveness of DL in predicting clinical outcomes for patients with major artery blockage from pretreatment diffusion-weighted image data [54].The convolutional neural network-based deep learning model used encoder-decoder architecture to segregate ischemic lesions, this model automatically extracted high-level feature maps in its middle layers and used their data to predict the clinical outcome. The deep learning model was able to extract more predictive data from pretreatment neuroimaging data than the conventional neuroimaging biomarkers.

Moyamoya Disease

Akiyama Y et al. used a deep-learning algorithm on T-2 weighted MRI images for diagnosing moyamoya disease. The algorithm was able to distinguish between moyamoya disease, atherosclerosis, and control group in basal cistern, basal ganglia, and centrum semiovale with an accuracy of 92.8%, 84.8%, and 87.8% respectively [55]. Chen Z et al. utilized six different machine learning methods to build the models and XGBoost, Logistic Regression (LR), and Support vector machine (SVM) models were used to determine risk of hemorrhage in patients with moyamoya disease. XGBoost model had the greatest area under the curve (0.874) and was determined to be a potential model for efficiently analyzing risk of hemorrhage [56].

Artery Stenosis

Machine learning models were developed and validated by Yeh CY et al. for classifying stenosis based on hemodynamic features. For Extracranial Carotid Doppler (ECCD), the random forest model performed better compared to the logistic regression model, with the predictive accuracy of the former being between 0.85 and 1. For Transcranial Doppler (TCD), the accuracy of random forest models to predict stenosis ranged from 0.67 to 0.86. The study thus indicated that machine learning-based models accurately classify artery stenosis [57].

Risk of Bias:The quality assessment of the findings of the included studies is summarised in figures 2 and 3.

Based on the ROBINS-I assessment, most studies appear to have a “low” risk of bias, with most of the domains receiving the maximum number of stars. However, as with any assessment tool, this is a rough adaptation of the ROBINS-I for non-randomised studies of interventions, and a more detailed assessment may uncover additional sources of bias.

This systematic review aimed to shed light on the various applications of AI in cerebrovascular surgery. AI has been progressively becoming significant in addressing diverse cerebrovascular disorders such as stroke, aneurysms, arteriovenous malformations etc. by enhancing diagnosis, risk stratification, predicting outcomes.

Predicting Clinical Outcomes:

Studies assessing the role of AI & ML models in AIS have shown a beneficial effect in predicting clinical outcomes such as time to treatment, door to skin puncture time, likelihood of home discharge, first pass reperfusion and feature analysis in patients with acute ischemic stroke [10], [12], [17], [22]. Notably, AI models have demonstrated higher accuracy as compared to conventional logistic regression model in prediction of first pass reperfusion and feature analysis [12]. The utilisation of CART models which incorporates fundamental data, functional parameters, and environmental factors has yielded precise predictions regarding the likelihood of home discharge following inpatient rehabilitation [22]. Although only a few have delved into its potential for predicting stroke recurrence, the available evidence has shown favourable outcomes [14]. Furthermore, research has shown that convolutional neural network based model has outperformed conventional neuroimaging markers in predicting clinical outcomes of patients with major artery blockade/thrombosis [54]. Similarly, collective findings of various studies suggest that machine learning techniques can accurately predict hematoma expansion or delayed cerebral ischemia (DCI) in cases of hemorrhage [49], [52].

Detection and Diagnosis: Stroke - Swift and Precise:

In the array of cerebrovascular surgical applications, AI predominantly found its application in analysing neuroradiological data for diagnosis, segmentation & detection by means of different models. This is particularly notable in time critical scenarios such as stroke, making it first of its applications to be integrated into actual clinical practice. It has been found to significantly reduce time to treatment and door to puncture time improving the outcomes for patients with LVO AIS [10]. The high accuracy metrics support AI as an adjunct tool in stroke diagnostics [16], [24]. Machine learning models have been developed and shown promising outcomes in classifying stenosis, diagnosing moyamoya disease, brain arteriovenous malformations, cerebrovascular thrombosis [53], [55], [57], [58]. Various neural networks can aid in accurate and quick diagnosis & segmentation of cerebral aneurysms, a task which can be labour intensive if done manually [21], [31]. Its use has also been explored in the treatment and detection of periprocedural complications of ICA [38], [46], Fig. 4.

Enhanced Risk Stratification:

Likewise, by leveraging advanced computational abilities of AI, neurosurgeons can make more informed decisions in assessing patient risk. Studies comparing ML models with traditional scoring systems have demonstrated higher accuracy of ML models in predicting risk of cerebral infarction in AF patients [7]. Similarly, AI models have outperformed both traditional LR model and scoring systems in foreseeing the risk of rupture in intracerebral aneurysms, thereby presenting the potential to tailor treatment approaches [39], [40], [41], [42], [44], [45], [47]. In case of moyamoya disease, evidence suggests the AI model can be a potential model for efficiently analyzing risk of hemorrhage [56].

Challenges and Considerations:

Despite the promising outlook, further research is required to better understand the implications of AI. One challenge stems from data limitation as acquiring large amounts of adequately categorised data to train ML/DL models can be difficult given patient privacy considerations. Moreover, the capabilities and results of AI models should be used to assist and enhance the efforts of clinicians. Therefore, more studies should focus on evaluating the efficacy of collaborative approach involving both AI and clinicians rather than simply comparing AI against clinicians.

AI in Resource-Limited Settings:

By leveraging the advantages of AI, cerebrovascular surgery can become more accessible and safer in resource-limited settings. It can reduce the need for costly interventions by predicting post-op complications and adverse events. Despite magnetic resonance imaging (MRI’s) being the mainstay of the diagnostic procedure for brain tumours, it remains very difficult for doctors to differentiate primary lesions from metastatic ones. Even when there is a suspicion of primary cancer and imaging suggests a non-metastatic lesion, 89% still turn out to be metastasis on biopsy [59]. Hence the integration of AI in cerebrovascular surgery will aid in making faster and accurate diagnosis, especially in regions where neuroradiologists are scarce, and invasive methods of testing are unavailable.

Integration into Surgical Workflow

Patel and coworkers showed that ML models could predict MCA territory stroke with an accuracy of 74.2% [15]. Immediate and efficient diagnosis of an acute ischemic stroke is essential in patient care. Mechanical thrombectomy is the gold standard treatment for acute ischemic strokes [60]. Criteria to be fulfilled before initiating treatment includes, confirmation of diagnosis and treatment prep within 6 hours of onset of symptoms. Similarly, the definitive treatment for hemorrhagic strokes include craniotomy, decompressive craniectomy, stereotactic aspiration, endoscopic aspiration, and catheter aspiration [61]. The STICH trial showed that selective patients have better outcomes with early surgery. Tang et. al. used ML models to accurately classify hematomas with more than a 90% accuracy rate, which is far superior than existing diagnostic frameworks [49]. AI not only helps diagnose stroke patients early, but also helps guide the flow of treatment. Matsoukas et al. concluded that usage of ML powered stroke triage software, such as Viz LVO, had a significant impact on patient outcome [24].

Chen Z et al developed a model to efficiently analyze the risk of hemorrhage in Moyamoya Disease (MMD), which represents a prong of its two prong dynamic pathophysiology [56]. The treatment of choice to circumevent this pathology is a superficial temporal artery-middle cerebral artery (STA-MCA) bypass (EC-IC bypass) [62]. With a better understanding of the underlying pathology, treatment protocols can be instituted earlier.

Intracranial aneurysm(IA) is a leading cause of stroke, whose management is almost purely surgical [63]. IAs are prone to rupture hence immediate treatment is recommended for almost all types of IAs. With several surgical modalities available for treatment - Simple Clipping, Temporal Artery Occlusion Technique, Bypass techniques, Simple Coiling, Balloon-Assisted Coiling, etc. each with their own benefits and drawbacks, new and better techniques are always welcome. Liu C. et al. developed an AI algorithm to automatically generate patient specific 3D microcatheters which showed successful usage, with minimal risk of perioperative complications [38].

Human Collaboration vs. Machine Dominance

By itself, big data - AI, ML, DL dont mean anything [5]. Especially in the context of such high stakes decision making. Relying solely on AI models can backfire in two ways. Models might either cling to meaningless patterns ("overfitting") or generate shaky results due to overlapping predictors ("multicollinearity"). The best-worst outcome is that both can inflate accuracy estimates and overpromise real-world effectiveness. Hence the chief usefulness of AI comes in its ability to transform data into knowledge, which in the correct hands can mean the difference between life and death. Models which can diagnose diseases based on imaging, quicker and more accurately than doctors will lead to better outcomes while performing time sensitive procedures. AI will also play an important role in organizing treatment protocols for diseases with a long and complicated treatment algorithm. It will ensure that the correct point of contact is notified instantly based on findings. Human collaboration with AI leads to an unique situation wherein the sum of the parts is greater than the whole.

Limitations

The transformative potential of AI in cerebrovascular surgery is undeniable, promising enhanced precision, improved decision-making, and optimized patient outcomes. However, realizing this vision requires rigorous further research to address current limitations and validate the clinical utility of AI-powered tools. A major challenge is due to rigid regulatory platforms. Any AI model will have a non 100% accuracy in diagnosing, triaging, recommending an ideal treatment plan for a patient. This raises concerns in the real-world especially for high-risk procedures such as those discussed above [3]. The second challenge lies in data acquisition. Training robust Machine Learning (ML) and Deep Learning (DL) models necessitates large volumes of accurately labeled cerebrovascular data. Unfortunately, stringent patient privacy regulations and the inherent complexity of vascular disease diagnosis can hinder data collection, impeding model development and generalization. Furthermore, the application of AI in the cerebrovascular operating room remains in its infancy.

Future Prospects

While theoretical benefits abound for AI models, translating them into tangible improvements for patients necessitates a deeper understanding of how AI can seamlessly integrate and augment the workflow of surgeons. Comprehensive evaluations are crucial to assess the true capabilities and impact of AI models in supporting and enhancing clinical decision-making during surgery. Exploring the potential of AI beyond the immediate surgical intervention presents further avenues for investigation. Research into pre-operative planning, risk stratification, and long-term patient monitoring could unlock additional benefits, paving the way for a more holistic and AI-driven approach to cerebrovascular care. A good example is the use of ML for COVID-19 diagnosis and work up. With proper resource management and focused effort of the worldwide scientific body, models were developed with a base accuracy of 99+% in diagnosing the disease [64]. This simply highlights the ceiling AI can achieve in favourable conditions. While the future of AI in cerebrovascular surgery appears bright, the journey to clinical integration demands a robust research agenda. It cannot solely be a single industry effort. Government regulators, journals, academics, and clinicians from around the globe need to arrive at an universal consensus and recognise the need and immense benefit AI provides in guiding high risk procedures in cerebrovascular surgery. Overcoming data limitations, delving deeper into the practical benefits of AI-assisted surgery, and extending its reach beyond the operating room are critical steps towards transforming cerebrovascular care and unlocking the full potential of this transformative technology.

This work aimed to offer a comprehensive review in the use of AI for cerebrovascular surgery. The existing models have improved outcomes by aiding clinicians and providing accurate, efficient, analyses of a variety of clinical predictors. One of the most promising fields AI has been used in is the preoperative diagnostic and risk stratification, particularly notable for diseases with a very time critical prognosis. Significant advances have been made in the diagnosis of acute ischemic strokes, and intracranial hemorrhages. AI has deeply affected the field of cerebrovascular surgery, with branches in disease detection and diagnosis, predicting clinical outcomes, risk stratification, aiding cerebrovascular surgery, and helping desiging new surgical procedures which were previously impossible. Substantial limitations still remain, which due to the nature of application, leaves very little room for mistakes. Hence it is essential that models are tested on truly independent validation data sets. It will require comprehensive collaboration between the regulatory bodies and academics to achieve this. Considering the progress we've already made, and how patient outcomes are changing in the real world courtesy AI models, it is highly likely that sooner rather than later AI will truly revolutionise the field of cerebrovascular surgery.

Ethics approval: This study does not require ethics approval.

Acknowledgement: None

Consent to participate: Not applicable

Funding: None. This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declarations of interest: None

Data availability: Not applicable

R. F. Spetzler et al., “The Barrow Ruptured Aneurysm Trial: 6-year results,” J Neurosurg, vol. 123, no. 3, pp. 609–617, Sep. 2015, doi: 10.3171/2014.9.JNS141749.
M. T. Lawton and G. E. Vates, “Subarachnoid Hemorrhage,” N Engl J Med, vol. 377, no. 3, pp. 257–266, Jul. 2017, doi: 10.1056/NEJMcp1605827.
F. Jiang et al., “Artificial intelligence in healthcare: past, present and future,” Stroke Vasc Neurol, vol. 2, no. 4, Dec. 2017, doi: 10.1136/svn-2017-000101.
J. T. Senders et al., “Machine Learning and Neurosurgical Outcome Prediction: A Systematic Review,” World Neurosurg, vol. 109, pp. 476-486.e1, Jan. 2018, doi: 10.1016/j.wneu.2017.09.149.
Z. Obermeyer and E. J. Emanuel, “Predicting the Future — Big Data, Machine Learning, and Clinical Medicine,” N Engl J Med, vol. 375, no. 13, pp. 1216–1219, Sep. 2016, doi: 10.1056/NEJMp1606181.
“Cochrane Handbook for Systematic Reviews of Interventions.” Accessed: May 09, 2024. [Online]. Available: https://handbook-5-1.cochrane.org/
H. Nishi et al., “Predicting cerebral infarction in patients with atrial fibrillation using machine learning: The Fushimi AF registry,” J Cereb Blood Flow Metab, vol. 42, no. 5, pp. 746–756, May 2022, doi: 10.1177/0271678X211063802.
A. Yahav-Dovrat et al., “Evaluation of Artificial Intelligence-Powered Identification of Large-Vessel Occlusions in a Comprehensive Stroke Center,” AJNR Am J Neuroradiol, vol. 42, no. 2, pp. 247–254, Jan. 2021, doi: 10.3174/ajnr.A6923.
Y. Bai, F. Liu, and H. Zhang, “Artificial Intelligence Limb Rehabilitation System on Account of Virtual Reality Technology on Long-Term Health Management of Stroke Patients in the Context of the Internet,” Comput Math Methods Med, vol. 2022, p. 2688003, 2022, doi: 10.1155/2022/2688003.
J. R. Morey et al., “Real-World Experience with Artificial Intelligence-Based Triage in Transferred Large Vessel Occlusion Stroke Patients,” Cerebrovasc Dis, vol. 50, no. 4, pp. 450–455, 2021, doi: 10.1159/000515320.
D. Park et al., “Machine learning predictive model for aspiration screening in hospitalized patients with acute stroke,” Sci Rep, vol. 13, no. 1, p. 7835, May 2023, doi: 10.1038/s41598-023-34999-8.
L. Velagapudi et al., “A Machine Learning Approach to First Pass Reperfusion in Mechanical Thrombectomy: Prediction and Feature Analysis,” J Stroke Cerebrovasc Dis, vol. 30, no. 7, p. 105796, Jul. 2021, doi: 10.1016/j.jstrokecerebrovasdis.2021.105796.
S. A. Sheth et al., “Machine Learning-Enabled Automated Determination of Acute Ischemic Core From Computed Tomography Angiography,” Stroke, vol. 50, no. 11, pp. 3093–3100, Nov. 2019, doi: 10.1161/STROKEAHA.119.026189.
K. Wang et al., “A machine learning model for visualization and dynamic clinical prediction of stroke recurrence in acute ischemic stroke patients: A real-world retrospective study,” Front Neurosci, vol. 17, p. 1130831, 2023, doi: 10.3389/fnins.2023.1130831.
T. R. Patel et al., “Revascularization Outcome Prediction for A Direct Aspiration-First Pass Technique (ADAPT) from Pre-Treatment Imaging and Machine Learning,” Brain Sci, vol. 11, no. 10, p. 1321, Oct. 2021, doi: 10.3390/brainsci11101321.
M. Livne et al., “A U-Net Deep Learning Framework for High Performance Vessel Segmentation in Patients With Cerebrovascular Disease,” Front Neurosci, vol. 13, p. 97, 2019, doi: 10.3389/fnins.2019.00097.
H. Nishi et al., “Predicting Clinical Outcomes of Large Vessel Occlusion Before Mechanical Thrombectomy Using Machine Learning,” Stroke, vol. 50, no. 9, pp. 2379–2388, Sep. 2019, doi: 10.1161/STROKEAHA.119.025411.
Y. Chung et al., “Label-free histological analysis of retrieved thrombi in acute ischemic stroke using optical diffraction tomography and deep learning,” J Biophotonics, vol. 16, no. 8, p. e202300067, Aug. 2023, doi: 10.1002/jbio.202300067.
J. You et al., “3D dissimilar-siamese-u-net for hyperdense Middle cerebral artery sign segmentation,” Comput Med Imaging Graph, vol. 90, p. 101898, Jun. 2021, doi: 10.1016/j.compmedimag.2021.101898.
K. Wang et al., “The predictors of death within 1 year in acute ischemic stroke patients based on machine learning,” Front Neurol, vol. 14, p. 1092534, 2023, doi: 10.3389/fneur.2023.1092534.
J. Liao et al., “Using a Convolutional Neural Network and Convolutional Long Short-term Memory to Automatically Detect Aneurysms on 2D Digital Subtraction Angiography Images: Framework Development and Validation,” JMIR Med Inform, vol. 10, no. 3, p. e28880, Mar. 2022, doi: 10.2196/28880.
T. Imura et al., “Decision Tree Algorithm Identifies Stroke Patients Likely Discharge Home After Rehabilitation Using Functional and Environmental Predictors,” J Stroke Cerebrovasc Dis, vol. 30, no. 4, p. 105636, Apr. 2021, doi: 10.1016/j.jstrokecerebrovasdis.2021.105636.
M. Bretzner et al., “Radiomics-Derived Brain Age Predicts Functional Outcome After Acute Ischemic Stroke,” Neurology, vol. 100, no. 8, pp. e822–e833, Feb. 2023, doi: 10.1212/WNL.0000000000201596.
S. Matsoukas et al., “AI software detection of large vessel occlusion stroke on CT angiography: a real-world prospective diagnostic test accuracy study,” J Neurointerv Surg, vol. 15, no. 1, pp. 52–56, Jan. 2023, doi: 10.1136/neurintsurg-2021-018391.
P.-J. Kim et al., “Another Look at Obesity Paradox in Acute Ischemic Stroke: Association Rule Mining,” J Pers Med, vol. 12, no. 1, p. 16, Dec. 2021, doi: 10.3390/jpm12010016.
J. Lyu et al., “Collateral-Core Ratio as a Novel Predictor of Clinical Outcomes in Acute Ischemic Stroke,” Transl Stroke Res, vol. 14, no. 1, pp. 73–82, Feb. 2023, doi: 10.1007/s12975-022-01066-9.
C. Ou et al., “A deep learning-based automatic system for intracranial aneurysms diagnosis on three-dimensional digital subtraction angiographic images,” Med Phys, vol. 49, no. 11, pp. 7038–7053, Nov. 2022, doi: 10.1002/mp.15846.
C. Ou, W. Chong, C.-Z. Duan, X. Zhang, M. Morgan, and Y. Qian, “A preliminary investigation of radiomics differences between ruptured and unruptured intracranial aneurysms,” Eur Radiol, vol. 31, no. 5, pp. 2716–2725, May 2021, doi: 10.1007/s00330-020-07325-3.
J. Wang et al., “Detection of Intracranial Aneurysms Using Multiphase CT Angiography with a Deep Learning Model,” Acad Radiol, vol. 30, no. 11, pp. 2477–2486, Nov. 2023, doi: 10.1016/j.acra.2022.12.043.
X. Liu et al., “Deep Learning for Detection of Intracranial Aneurysms from Computed Tomography Angiography Images,” J Digit Imaging, vol. 36, no. 1, pp. 114–123, Feb. 2023, doi: 10.1007/s10278-022-00698-5.
H. Duan, Y. Huang, L. Liu, H. Dai, L. Chen, and L. Zhou, “Automatic detection on intracranial aneurysm from digital subtraction angiography with cascade convolutional neural networks,” BioMedical Engineering OnLine, vol. 18, no. 1, p. 110, Nov. 2019, doi: 10.1186/s12938-019-0726-2.
H. Jin et al., “Fully automated intracranial aneurysm detection and segmentation from digital subtraction angiography series using an end-to-end spatiotemporal deep neural network,” J Neurointerv Surg, vol. 12, no. 10, pp. 1023–1027, Oct. 2020, doi: 10.1136/neurintsurg-2020-015824.
L. Benvenuti, S. Chibbaro, S. Carnesecchi, F. Pulerà, and R. Gagliardi, “Automated three-dimensional volume rendering of helical computed tomographic angiography for aneurysms: an advanced application of neuronavigation technology,” Neurosurgery, vol. 57, no. 1 Suppl, pp. 69–77; discussion 69-77, Jul. 2005, doi: 10.1227/01.neu.0000163485.56639.7e.
W. You et al., “Protocol and Preliminary Results of the Establishment of Intracranial Aneurysm Database for Artificial Intelligence Application Based on CTA Images,” Front Neurol, vol. 13, p. 932933, 2022, doi: 10.3389/fneur.2022.932933.
K. E. Poppenberg et al., “Classification models using circulating neutrophil transcripts can detect unruptured intracranial aneurysm,” J Transl Med, vol. 18, no. 1, p. 392, Oct. 2020, doi: 10.1186/s12967-020-02550-2.
J. Su et al., “Diagnosis of Unruptured Intracranial Aneurysm by High-Performance Serum Metabolic Fingerprints,” Small Methods, vol. 7, no. 3, p. e2201486, Mar. 2023, doi: 10.1002/smtd.202201486.
S. S. Veeturi et al., “Hemodynamic Analysis Shows High Wall Shear Stress Is Associated with Intraoperatively Observed Thin Wall Regions of Intracranial Aneurysms,” Journal of Cardiovascular Development and Disease, vol. 9, no. 12, Art. no. 12, Dec. 2022, doi: 10.3390/jcdd9120424.
C. Liu, Y. Shen, X. Wu, K. Qian, X. Hu, and H. Yang, “Artificial Intelligence-Assisted Microcatheter Shaping for Intracranial Aneurysm Coiling: A Preliminary Study,” Ann Vasc Surg, vol. 85, pp. 228–236, Sep. 2022, doi: 10.1016/j.avsg.2022.03.013.
J. Heo, S. J. Park, S.-H. Kang, C. W. Oh, J. S. Bang, and T. Kim, “Prediction of Intracranial Aneurysm Risk using Machine Learning,” Sci Rep, vol. 10, p. 6921, Apr. 2020, doi: 10.1038/s41598-020-63906-8.
C. Ou et al., “Automated Machine Learning Model Development for Intracranial Aneurysm Treatment Outcome Prediction: A Feasibility Study,” Front. Neurol., vol. 12, Nov. 2021, doi: 10.3389/fneur.2021.735142.
P. Li et al., “A deep-learning method for the end-to-end prediction of intracranial aneurysm rupture risk,” Patterns (N Y), vol. 4, no. 4, p. 100709, Apr. 2023, doi: 10.1016/j.patter.2023.100709.
J. Liu et al., “Prediction of rupture risk in anterior communicating artery aneurysms with a feed-forward artificial neural network,” Eur Radiol, vol. 28, no. 8, pp. 3268–3275, Aug. 2018, doi: 10.1007/s00330-017-5300-3.
M. M. Shiraz Bhurwani et al., “Feasibility study for use of angiographic parametric imaging and deep neural networks for intracranial aneurysm occlusion prediction,” J Neurointerv Surg, vol. 12, no. 7, pp. 714–719, Jul. 2020, doi: 10.1136/neurintsurg-2019-015544.
X. Tong et al., “Rupture discrimination of multiple small (< 7 mm) intracranial aneurysms based on machine learning-based cluster analysis,” BMC Neurol, vol. 23, p. 45, Jan. 2023, doi: 10.1186/s12883-023-03088-8.
T. Lu et al., “A machine learning-derived gene signature for assessing rupture risk and circulatory immunopathologic landscape in patients with intracranial aneurysms,” Front Cardiovasc Med, vol. 10, p. 1075584, 2023, doi: 10.3389/fcvm.2023.1075584.
Z. Tian, W. Li, X. Feng, K. Sun, and C. Duan, “Prediction and analysis of periprocedural complications associated with endovascular treatment for unruptured intracranial aneurysms using machine learning,” Front Neurol, vol. 13, p. 1027557, Oct. 2022, doi: 10.3389/fneur.2022.1027557.
C. Ou et al., “Rupture Risk Assessment for Cerebral Aneurysm Using Interpretable Machine Learning on Multidimensional Data,” Front Neurol, vol. 11, p. 570181, Dec. 2020, doi: 10.3389/fneur.2020.570181.
J. Nawabi et al., “Imaging-Based Outcome Prediction of Acute Intracerebral Hemorrhage,” Transl Stroke Res, vol. 12, no. 6, pp. 958–967, Dec. 2021, doi: 10.1007/s12975-021-00891-8.
Z. Tang et al., “Deep Learning-Based Prediction of Hematoma Expansion Using a Single Brain Computed Tomographic Slice in Patients With Spontaneous Intracerebral Hemorrhages,” World Neurosurg, vol. 165, pp. e128–e136, Sep. 2022, doi: 10.1016/j.wneu.2022.05.109.
J. J. Park, C. Kim, and J. P. Jeon, “Monitoring of Delayed Cerebral Ischemia in Patients with Subarachnoid Hemorrhage via Near-Infrared Spectroscopy,” J Clin Med, vol. 9, no. 5, p. 1595, May 2020, doi: 10.3390/jcm9051595.
S. Rangaraj et al., “Identifying risk factors of intracerebral hemorrhage stability using explainable attention model,” Med Biol Eng Comput, vol. 60, no. 2, pp. 337–348, Feb. 2022, doi: 10.1007/s11517-021-02459-y.
P. Hu et al., “Comparison of Conventional Logistic Regression and Machine Learning Methods for Predicting Delayed Cerebral Ischemia After Aneurysmal Subarachnoid Hemorrhage: A Multicentric Observational Cohort Study,” Front Aging Neurosci, vol. 14, p. 857521, 2022, doi: 10.3389/fnagi.2022.857521.
J.-S. Hong et al., “Deep Learning Detection and Segmentation of Brain Arteriovenous Malformation on Magnetic Resonance Angiography,” J Magn Reson Imaging, vol. 59, no. 2, pp. 587–598, Feb. 2024, doi: 10.1002/jmri.28795.
H. Nishi et al., “Deep Learning-Derived High-Level Neuroimaging Features Predict Clinical Outcomes for Large Vessel Occlusion,” Stroke, vol. 51, no. 5, pp. 1484–1492, May 2020, doi: 10.1161/STROKEAHA.119.028101.
Y. Akiyama, T. Mikami, and N. Mikuni, “Deep Learning-Based Approach for the Diagnosis of Moyamoya Disease,” J Stroke Cerebrovasc Dis, vol. 29, no. 12, p. 105322, Dec. 2020, doi: 10.1016/j.jstrokecerebrovasdis.2020.105322.
Z. Chen, H. Luo, and L. Xu, “Machine learning models of ischemia/hemorrhage in moyamoya disease and analysis of its risk factors,” Clin Neurol Neurosurg, vol. 209, p. 106919, Oct. 2021, doi: 10.1016/j.clineuro.2021.106919.
C.-Y. Yeh et al., “Development and Validation of Machine Learning Models to Classify Artery Stenosis for Automated Generating Ultrasound Report,” Diagnostics, vol. 12, no. 12, Art. no. 12, Dec. 2022, doi: 10.3390/diagnostics12123047.
X. Yang et al., “Deep Learning Algorithm Enables Cerebral Venous Thrombosis Detection With Routine Brain Magnetic Resonance Imaging,” Stroke, vol. 54, no. 5, pp. 1357–1366, May 2023, doi: 10.1161/STROKEAHA.122.041520.
Naidu, Bodepu Sai, et al. “Scenic: An area and energy-efficient CNN-based hardware accelerator for discernable classification of brain pathologies using MRI.” 2022 35th International Conference on VLSI Design and 2022 21st International Conference on Embedded Systems (VLSID), Feb. 2022, https://doi.org/10.1109/vlsid2022.2022.00042.
W. J. Powers et al., “Guidelines for the Early Management of Patients With Acute Ischemic Stroke: 2019 Update to the 2018 Guidelines for the Early Management of Acute Ischemic Stroke: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke Association,” Stroke, vol. 50, no. 12, pp. e344–e418, Dec. 2019, doi: 10.1161/STR.0000000000000211.
S. Ojaghihaghighi, S. S. Vahdati, A. Mikaeilpour, and A. Ramouz, “Comparison of neurological clinical manifestation in patients with hemorrhagic and ischemic stroke,” World J Emerg Med, vol. 8, no. 1, pp. 34–38, 2017, doi: 10.5847/wjem.j.1920-8642.2017.01.006.
T. Tominaga et al., “Recommendations for the Management of Moyamoya Disease: A Statement from Research Committee on Spontaneous Occlusion of the Circle of Willis (Moyamoya Disease) [2nd Edition],” Surgery for Cerebral Stroke, vol. 46, no. 1, pp. 1–24, 2018, doi: 10.2335/scs.46.1.
J. Zhao, H. Lin, R. Summers, M. Yang, B. G. Cousins, and J. Tsui, “Current Treatment Strategies for Intracranial Aneurysms: An Overview,” Angiology, vol. 69, no. 1, pp. 17–30, Jan. 2018, doi: 10.1177/0003319717700503.
S. Biswas, S. Chatterjee, A. Majee, S. Sen, F. Schwenker, and R. Sarkar, “Prediction of COVID-19 from Chest CT Images Using an Ensemble of Deep Learning Models,” Applied Sciences, vol. 11, no. 15, Art. no. 15, Jan. 2021, doi: 10.3390/app11157004.

Table 1: PICO Criteria for Study Selection

Criteria	Inclusion	Exclusion
Population	People diagnosed with cerebrovascular disorders requiring neurosurgical interventions	- Studies not published in English - Not relevant to study - Full text not available - Low screening score
Intervention	Studies investigating the use of artificial intelligence techniques or applications in cerebrovascular neurosurgery	- -
Comparators	Other treatment modalities	- -
Study designs	Original Studies, Observational Studies, Case report, Case series	- Narrative Reviews, Editorials, Short communications, Case studies, Systematic review and meta-analysis

Table 2. Detailed search strategy for PubMed, SCOPUS, WoS.

Database	Search Strategy	Record Identified(n)
Pubmed	("Artificial Intelligence" OR "Machine Learning" OR "Deep Learning" OR "Computer Vision") AND ("Stroke" OR "Intracranial Aneurysm" OR "Arteriovenous Malformations" OR "Cerebral Ischemia") AND ("Neurosurgery" OR "Neurosurgical Procedures") AND ("Innovation" OR "Technological Innovations" OR "Technology Assessment, Biomedical" OR "Technology")	56
Scopus	("Artificial Intelligence" OR "Machine Learning" OR "Deep Learning" OR "Computer Vision") AND ("Stroke" OR "Intracranial Aneurysm" OR "Arteriovenous Malformations" OR "Cerebral Ischemia") AND ("Neurosurgery" OR "Neurosurgical Procedures") AND ("Innovation" OR "Technological Innovations" OR "Technology Assessment, Biomedical" OR "Technology")	73
Web of Sciences	("Artificial Intelligence" OR "Machine Learning" OR "Deep Learning" OR "Computer Vision") AND ("Stroke" OR "Intracranial Aneurysm" OR "Arteriovenous Malformations" OR "Cerebral Ischemia") AND ("Neurosurgery" OR "Neurosurgical Procedures") AND ("Innovation" OR "Technological Innovations" OR "Technology Assessment, Biomedical" OR "Technology")	34

Table 3: AI Integration in Acute Ischemic Stroke Management: An Overview of Included Studies

Variety

First author

Year

Journal

Location

Study design

Sample size

AI/ML used

Results

Risk prediction

Hidehisa Nishi

2022

Journal of cerebral blood flow and metabolism :

United States

Prospective cohort

1757

Gradient boosting tree algorithm

The authors reported that their machine learning model had a higher predictive accuracy than the CHADS2 and CHA2DS2-VASc risk scores

Detection

A Yavhat-Dovrat

2021

AJNR. American journal of neuroradiology

United States

Retrospective study

1167 CTAs, 404

stroke protocols

Viz LVO Algorithm (v14.3)

Sensitivity was 0.82

(95% CI, 0.71–0.89),

The system showed suboptimal sensitivity, which prevents it from being used as a diagnostic tool to date

Outcome prediction

Yu Bai

2022

Computational and mathematical methods in medicine

United States

Randomized control study

VR digital training

The system had a better effect on the prognosis of stroke patients than ordinary training and had a high clinical value. Although the experiment proves that this method has a high rehabilitation effect, it was still recommended to wait for the development of virtual reality technology

Outcome prediction

Jacob R. Morey

2021

Cerebrovascular diseases (Basel, Switzerland)

Switzerland

Retrospective Analysis

Viz LVO Algorithm

The median initial doorto-neuroendovascu lar team (NT) notification time interval was significantly faster (25.0 min [IQR = 12.0]

vs. 40.0 min [IQR = 61.0]; p = 0.01) with

less variation (p < 0.05) following Viz LVO implementation. The median initial

door-to-skin puncture time interval was 25 min shorter in the post-Viz cohort, although this was not statistically significant (p = 0.15).

Outcome prediction

Dougho Park

2023

Scientific reports

England

Retrospective study

3408

RLRs, RF, XGB, NB, SVM

A videofluoroscopic swallowing study (VFSS) confirmed aspiration, and the results showed that the proposed machine learning prediction models are valid and practical for screening aspiration in patients with acute stroke

Detection

Lohit Velagapudi

2021

Journal of stroke and cerebrovascular diseases : the official journal of National Stroke Association

United States

Retrospective Study

220

e scikit-learn package for Python

Accuracy for the Random Forest and SVM models were 67.1% compared to 65.8% for the logistic regression model. Brier score was lower for the Random Forest model (0.329 vs 0.342)

indicating better predictive capability

Risk prediction

Sunil A Sheth

2019

Stroke

United States

Retrospective Study

297

DeepSymNet

The DeepSymNet algorithm autonomously learned to identify the intracerebral vasculature on CTA and detected LVO with AUC 0.88. The method was also able to determine infarct core as defined by

CTP-RAPID from the CTA source images with AUC 0.88 and 0.90

Outcome prediction

Kai Wung

2023

Frontiers in neuroscience

Switzerland

Prospective Study

645

LR,NBC,DT,RF,XGB, GBM

The results showed that the RF model was the best predictor of stroke recurrence (mean AUC 0.959, standard

deviation 0.017) and the LR model had the lowest AUC value of 0.777. HCY has the greatest impact on the model, HCY, SS, and CRP all contribute positively to the RF model output prediction results, and SOH contributes negatively to the RF model output prediction results.

Detection

Tatsat R Patel

2021

Brain sciences

Switzerland

Retrospective Study

119

Multivariate ML

Perviousness (p = 0.002) and AOI (p =

0.031) were significantly higher and clot length (p = 0.007) was significantly lower in ADAPT cases with

FPE. A logistic regression model achieved the highest accuracy (74.2%) in the testing cohort, with an AUC = 0.769.

Outcome prediction

Michelle Libne

2019

Frontiers in neuroscience

Switzerland

Retrospective Study

U- Net CNN Model

The U-net models yielded high performance for both the full and the reduced architecture: The visual analysis revealed excellent performance in large vessels and sufficient performance in small vessels.

Risk prediction

Hidehisa nishi

2019

Stroke

United States

Retrospective Cohort Study

502

RLR,LSV,RF

The area under the receiver operating characteristic curve of random forest, which was the worst among the machine learning algorithms, was significantly higher than those of the standard statistical model and the best model among the previously reported pretreatment scoring methods in the derivation

Risk prediction

Yoonjae Chung

2023

Journal of biophotonics

Germany

Label free quantification

15 samples

ODT, DL

The DL model classifies ODT image patches with 95% accuracy, and the collective prediction generates a whole-slide map of red blood cells and fibrin.

Detection

Jia You

2021

Computerized medical imaging and graphics

United States

Retrospective Study

624

DSU-Net

The HMCAS was detected in 195 (31.25

%) patients. The HMCAS has a mean volume of 211.31 (standard deviation of 160.32 ) and a mean Hounsfield Unit (HU) of

40.15 ( standard deviation of 6.08) on a

per-patient basis

Risk prediction

Kai Wang

2023

Frontiers in neurology

Switzerland

Retrospective Study

677

LR,NBC,DT,RF,XGB, GBM

Multivariate logistic regression analysis revealed that C-reactive protein (CRP), homocysteine (HCY) levels, stroke severity (SS), and the number of stroke lesions (NOS) were independent risk factors for death within 1 year in patients with AIS. The area under the curve value of the XGB model was 0.846, which was the highest among the six ML algorithms.

Detection

JunHua Liao

2022

JMIR medical informatics

Canada

Case-control

275

C-LSTM framework

More spatial and temporal information can help improve the performance of the frameworks and,the

bi-input+RetinaNet+C-L STM framework had the best performance when compared to that of the other frameworks.The study demonstrates that this system can assist physicians in detecting intracranial aneurysms on 2D DSA images

Outcome prediction

Takeshi Imura

2021

Journal of stroke and cerebrovascular diseases : the official journal of National Stroke Association

United States

Retrospective study

1125

CART Model

The CART model incorporating basic information, functional factors, and environmental factors achieved the highest accuracy for identifying home discharge.

Outcome prediction

Martin Bretzner

2023

Neurology

United States

Retrospective study

4,163

ElasticNet linear regression model

Factors such as age, RBA, National Institutes of Health Stroke Scale (NIHSS) score, and a history of prior stroke were significantly associated with functional outcomes.

The negative effect of RBA on outcome was particularly notable in minor strokes.

Detection

Stavros Matsoukas

2023

Journal of neurointerventional surgery

England

Prospective Study

1822

Viz LVO

Detection rates for ICA-T, M1, and M2

occlusions were 100%,

93%, and 49%,

respectively. As expected, the algorithm offered better detection rates for proximal occlusions than for mid/distal M2 occlusions (58% vs

28%, p=0.03).

Risk Prediction

Pum-Jun Kim

2021

Journal of personalized medicine

Switzerland

Retrospective study

2580

Association Rule Mining

The binary logistic regression (BLR) analysis showed that being obese had a positive impact on the outcome. Additionally, the association rule mining (ARM) analysis revealed that obese patients with good outcomes were also associated with an age less than 55 years and mild stroke severity.

Outcome prediction

Jinhao Lyu

2023

Translational stroke research

China

Prospective study

129

XGB, SHAP

In 12 patients, integrated digital subtraction angiography and automated 3-D volume-rendered images allowed an accurate presurgical evaluation.

Furthermore, in all patients on whom surgery was performed, aneurysms were found in the exact location and with the same anatomic features as depicted by the 3-D volume-rendered models.

Table 4: AI Integration in Intracranial Aneurysm Management: An Overview of Included Studies

TABLE 2
Title	Year	Journal	Study design	Patient characteristics	Sample size	AI/ML model used	Purpose of use of AI/ML model	Outcome measured	Follow-up duration	Results
A deep learning-based automatic system for intracranial aneurysms diagnosis on three-dimensional digital subtraction angiographic images.	2022	Medical physics	Retrospective Study	intracranial aneurysms on digital subtraction angiography (DSA)	263 total	cross-scale dual-path transformer module	to develop a comprehensive system for automatic aneurysm detection, morphology measurement, and aneurysm location identification	performance of the proposed system for automatic aneurysm detection	Not provided	a deep learning-based automatic system for intracranial aneurysm diagnosis using a 3D convolutional neural network and a cross-scale dual-path transformer module. The proposed system achieved state-of-the-art detection performance, 100% accuracy of parent artery classification, and accurate morphology measurement at the same level as neuroradiologists. The system can be a potential tool to assist in the diagnosis and risk management of aneurysms.
Deep Learning for Detection of Intracranial Aneurysms from Computed Tomography Angiography Images.	2023	Journal of digital imaging	Retrospective Study	suspected untreated intracranial aneurysm who underwent a DSA or a clipping operation after the CTA	90 cases	(3D) CNN model implemented	reliably segment and detect intracranial aneurysms from CTA images with high sensitivity	(PPV), and false positives for the detection and segmentation of intracranial aneurysms from CTA images using AI	Not provided	a 3D convolutional neural network (CNN) system for the detection and segmentation of intracranial aneurysms from (CTA) images. it achieved high sensitivity and (PPV) for the detection and segmentation of intracranial aneurysms.With better results for large- and medium-sized aneurysms.
Detection of Intracranial Aneurysms Using Multiphase CT Angiography with a Deep Learning Model.	2023	Academic radiology	Retrospective Study	FULL TEXT UNAVAILABLE	1383 patients
Fully automated intracranial aneurysm detection and segmentation from digital subtraction angiography series using an end-to-end spatiotemporal deep neural network.	2020	Journal of neurointerventional surgery	Retrospective Study	Aneurysm Patients	347 for model development, 147 for accuracy testing	Deep Neural Network Based Framework	Fully automated detection and segmentation deep neural network based framework to assist neurologists in evaluating and contouring intracranial aneurysms from 2D+time digital subtraction angiography (DSA) sequences during diagnosis	Succesful detection of Aneurysms	No Follow Up	Of the 354 aneurysms, 316 (89.3%) were successfully detected, corresponding to a patient level sensitivity of 97.7% at an average false positive number of 3.77 per sequence. The system runs for less than one second per sequence with an average dice coefficient score of 0.533
A preliminary investigation of radiomics differences between ruptured and unruptured intracranial aneurysms.	2021	European radiology	Retrospective Study	Cerebral Aneurysm	122 aneurysm (93 unruptured, 29 ruptured)	PyRadiomics module in Python for extraction of radiomics features. 3 machine learning algorithms: 1.Generalised Logisctic Regression 2.LASSO Regression 3. Ridge Regression	To investigate whether radiomics features are useful in predicting aneurysm stability in addition to aneurysm morphology	Outcome prediction and risk assessment: by showing the significant differences in radiomics between ruptured and unruptured aneurysms.	No Follow Up	Five morphological features and 89 radiomics features were significantly associated with rupture. Model A, model B, and model C achieved the area under the receiver operating characteristic curve of 0.767, 0.807, and 0.879, respectively. Model C was significantly better than model A and model B (p < 0.001). Multivariate analysis identified two radiomics features which were used to construct the simplified model showing an AUROC of 0.876.
Classification models using circulating neutrophil transcripts can detect unruptured intracranial aneurysm.	2020	Journal of translational medicine	retrospective case-control study	patients with and without unruptured intracranial aneurysms (IA).	134 patients (55 with IA, 79 IA-free controls)	Least Absolute Shrinkage and Selection Operator (LASSO)	to improve the accuracy of predicting the presence of unruptured intracranial aneurysms (IA) using circulating neutrophil transcripts.	accuracy of predictive models for unruptured intracranial aneurysms (IA) using circulating neutrophil transcripts.	No Follow Up	The testing performance across all methods had an average area under ROC curve (AUC) = 0.97, an improvement over our previous models. The Random Forest model performed best across both training and testing cohorts. RT-qPCR confirmed expression differences in 7 of 9 genes tested. Gene ontology and IPA network analyses performed on the 37 model genes reflected dysregulated inflammation, cell signaling, and apoptosis processes.
Automated three-dimensional volume rendering of helical computed tomographic angiography for aneurysms: an advanced application of neuronavigation technology.	2005	Neurosurgery	Prospective	Intracranial Aneurysms	15	Three-dimensional (3-D) volume rendering with CT	Introduce volume rendered helical CT angiographic data sets to evaluate clinical usefulness of the method in planning and performing surgical treatment of intracranial aneurysms	Anatomical position of aneurysm, treatment and its outcome	No Follow Up	Because of the very short time necessary to complete this process (<5 min to obtain 3-D volume-rendered images), it was possible to perform emergency clipping of the aneurysms in two patients who had been admitted in very compromised neurological conditions. In 12 patients, integrated digital subtraction angiography and automated 3-D volume-rendered images allowed an accurate presurgical evaluation. Furthermore, in all patients on whom surgery was performed, aneurysms were found in the exact location and with the same anatomic features as depicted by the 3-D volume-rendered models.
Automatic Detection of Intracranial Aneurysm from Digital Subtraction Angiography with Cascade Networks	2019	BioMedical Engineering Online	Prospective	Intracranial Aneurysm	241 ( for training ) , 40 (testing)	two staged CNN (convolutional neural network) based architecture	Automatic detection of intracranial aneurysms on 2D-DSA images	Accuracy of detection and time cost	No Follow Up	accuracy of 93.5% and the AUC of 0.942;detection time cost 0.569s (which was one hundred times faster than the classical DIP method of 62.546 s)
Protocol and Preliminary Results of the Establishment of Intracranial Aneurysm Database for Artificial Intelligence Application Based on CTA Images.	2022	Frontiers in neurology	Retrospective Study	Aneurysm Patients	3190 cases with 4124 aneurysms for database. Tested on 2272 Head CTAs with 2938 Aneurysms	3D-Unet Algorithm	Establishment of intracranial aneurysm database for AI application based on CTA images	1. Aneurysm Detection and Segmentation 2. Risk Prediction for growth and rupture of an Aneurysm	No Follow Up	For the test dataset, the recall and FP/case of the model to detect aneurysms were 0.964 and 2.01. For the aneurysm sizes <3, 3–7, and >7 mm, the recall was 0.894, 0.983, and 0.982, respectively. The Dice similarity coefficient was used to evaluate the model performance of IA segmentation. For the test dataset, the Dice values are 0.783. For the aneurysm sizes <3, 3–7, and >7 mm, the Dice values are 0.635, 0.796, and 0.868, respectively
Diagnosis of Unruptured Intracranial Aneurysm by High-Performance Serum Metabolic Fingerprints.	2023	Small methods	Retrospective case-control study	patients with unruptured intracranial aneurysm	not mentioned	not mentioned	to develop a machine learning model for early diagnosis of unruptured intracranial aneurysm(UIA) using serum metabolic fingerprints.	diagnostic performance of the machine learning model for the early diagnosis of unruptured intracranial aneurysm (UIA) using serum metabolic fingerprints which was assessed using the area under the curve (AUC) metric.The study also aimed to identify a metabolic biomarker panel for the diagnosis of UIA's.	No Follow Up	machine learning model achieved an AUC of 0.94 in the diagnosis of unruptured intracranial aneurysm(UIA).Lactate, glutamine, homoarginine, and 3-methylglutaconic acid are identified as the metabolic biomarker panel, which showed satisfactory diagnosis (AUC of 0.812, 95% CI: 0.727-0.897) and effective growth risk assessment (p<0.05, two-tailed t-test) of UIAs.
Hemodynamic Analysis Shows High Wall Shear Stress Is Associated with Intraoperatively Observed Thin Wall Regions of Intracranial Aneurysms.	2022	Journal of cardiovascular development and disease	Retrospective study	intracranial aneurysms	27	support vector machines	for image analysis to classify different regions of the intracranial aneurysm (IA) wall based on their appearance in intraoperative imaging.	The study measured several hemodynamic parameters, including wall shear stress (WSS), normalized WSS (WSSNorm), WSS divergence, transverse WSS, and residence time (RRT). These parameters were compared among different regions of the IA wall (thin, normal, and thick).	No Follow Up	The study found that thin regions of the IA wall exhibited significantly higher WSS, WSSNorm, WSS divergence, and transverse WSS compared to normal and thick regions. Thicker regions tended to have higher RRT compared to thin regions. Regression analysis revealed a significant positive correlation between WSS and thin regions, as well as a significant negative correlation between WSS divergence and thick regions.

Title	Year	Journal	Study design	Patient characteristics	Sample size	AI/ML model used	Purpose of use of AI/ML model	Outcome measured	Follow-up duration	Results
Prediction of Intracranial Aneurysm Risk using Machine Learning.	2020	Scientific reports	Retrospective Cohort	individuals at risk of intracranial aneurysms	427,269	XGB, LR, DNN, and RF models	identifying subjects at high risk of an intracranial aneurysm	develope a prediction model that can aid future IA screening strategies.	3 years	The authors concluded that their machine learning model could be used to identify subjects at high risk of IA and provide adequate radiological screening guidelines and effectively allocate medical resources.
Rupture Risk Assessment for Cerebral Aneurysm Using Interpretable Machine Learning on Multidimensional Data.	2020	Frontiers in neurology	Retrospective cohort	Aneuryms without any other intracranial vascular malformations	374	Prediction model: VM, ANN, and extreme gradient boosting (XGBoost). Shapley Additive Explanations (SHAP) analysis was applied to improve the interpretability of the ML model.	Assessment of cerebral aneurysm rupture risk	Risk assessment and comparitive evaluation		ML is feasible to develop aneurysm rupture risk models using multi-aspects data obtained from patient demographics, clinical characteristics, lifestyle behaviors, lipid profiles, and angiographic images and performs better in comparison to conventional statistical models like LR
A deep-learning method for the end-to-end prediction of intracranial aneurysm rupture risk.	2023	Patterns (New York, N.Y.)	Retrospective study	Intracranial Aneurysm	556	TransIAR net, an end to end deep learning method	A multiscale 3D convolutional neural network (CNN) to extract the structural patterns of the IA and its neighborhood with a dual branch of shared network structures ; the spatial dependence within the IA neighborhood with a transformer encoder;	Comparitive evaulations in terms of patient outcomes	-	The features learned by TransIAR are more effective and robust than handcrafted features, resulting in a 10% 15% improvement in the accuracy of rupture status prediction.
Feasibility study for use of angiographic parametric imaging and deep neural networks for intracranial aneurysm occlusion prediction.	2020	Journal of neurointerventional surgery	Retrospective Study	Patients with intrcranial anuerysms treated only with PED Pipeline Embolization Device	190 IA	Deep Neural Network (DNN) and Angiographic Parametric Imaging (API)	To predict IA occlusions, and to study the ability of DNN to to predict the occlusion outcomes immediately after device employement without waiting for the follow up period	Prediction accuracies of DNN in identifying IA occlusions for normalised, un-normalised and relative subgroups to minimise variability	Cases considered needed a minimum follow-up duration of 6 months (MEAN 9 MONTHS) adter PED, however the ML model enabled identification of the AI occlusion immediately following PED insertion	For the un-normalized, normalized, and relative subgroups, the DNN average prediction accuracies for IA occlusion were 62.5% (95% CI 60.5% to 64.4%), 70.8% (95% CI 68.2% to 73.4%), and 77.9% (95% CI 76.2% to 79.6%). DNNs can use quantitative imaging information from API to predict IA occlusion immediately postdevice deployment instead of waiting a period of minimum 6 months post-treatment. However, each API parameter cannot be used individually to make an accurate IA occlusion prediction model and multiple parameters of statostical relevance must be considered together to enchance the accuracy of the prediction outcomes
Prediction of rupture risk in anterior communicating artery aneurysms with a feed-forward artificial neural network.	2018	European radiology	retrospective study	ACOM aneurysms	594	two-layer feed-forward artificial neural network (ANN)	to predict the risk of rupture in ACOM aneurysms.	13 morphological parameters of ACOM aneurysms obtained from CT angiography (CTA) images, two demographic factors, and histories of hypertension and smoking.	-	several parameters, including age, vessel size, aneurysm height, perpendicular height, aneurysm neck size, aspect ratio, size ratio, aneurysm angle, vessel angle, aneurysm projection, A1 segment configuration, aneurysm lobulations, and hypertension, were significantly different between the ruptured and unruptured groups. the ANN model showed good performance and could serve as a valuable tool for predicting rupture risk in ACOM aneurysms, potentially aiding in the management of unruptured ACOM aneurysms.
Automated Machine Learning Model Development for Intracranial Aneurysm Treatment Outcome Prediction: A Feasibility Study.	2021	Frontiers in neurology	Prospective cohort study	Aneurysm treated by endovascular treatment	182 Patients (218 aneurysms)	AutoML derived model	long term post operative outcome prediction in aneurysm treatment	Outcome prediction and comparitive evaluation	data collected 6 months post surgery with 1 year follow up	AutoML model performed better in outcome prediction than statistical and manually derived ML models
Rupture discrimination of multiple small (<â7Â mm) intracranial aneurysms based on machine learning-based cluster analysis.	2023	BMC neurology	Retrospective Study	Small multiple intracranial aneurysms (SMIAs)	1,427 SMIAs from 660 patients	Machine learning-based method, cluster analysis	To classify complex epidemiological factors and comorbidities into simple clusters (phenogroupings) that aids to (1) analyze patient-level risk factors to identify the clinical phenotypes that are most relevant for discriminating multi-factorial clusters among patients with MIAs; and (2) analyze aneurysm-level risk factors to evaluate the association between these clusters and aneurysm rupture.	Risk Assessment: An unsupervised machine learning-based algorithm successfully identifed three distinct clusters with diferent SAH ( Sub Arachnoid Hemorrhages) risk in patients with SMIAs.	-	Cluster 1 (n=45) had the highest risk of subarachnoid hemorrhage (SAH) (75.6%) and was characterized by a higher prevalence of familiar IAs. Cluster 2 (n=110) had a moderate risk of SAH (38.2%) and was characterized by the highest rate of SAH history and highest number of vascular risk factors. Cluster 3 (n=505) had a relatively mild risk of SAH (17.6%) and was characterized by a lower prevalence of SAH history and lower number of vascular risk factors. Lasso regression analysis showed that compared with cluster 3, clusters 1 (odds ratio [OR], 7.391; 95% confdence interval [CI], 4.074–13.150) and 2 (OR, 3.014; 95% CI, 1.827–4.970) were at a higher risk of aneurysm rupture. In terms of performance, the area under the curve of the model was 0.828 (95% CI, 0.770–0.833).
A machine learning-derived gene signature for assessing rupture risk and circulatory immunopathologic landscape in patients with intracranial aneurysms.	2023	Frontiers in cardiovascular medicine	Experimental	Patients with ruptured and Unruptured intracranial Anueryms diagnosed by DSA	28 blood samples	MLDGS model with integrated random forest (RF), support vector machine (SVM), and least absolute shrinkage and selection operator (LASSO) regression and 12 algorithm models	to develop a molecular assay tool for optimizing the IA risk monitoring system.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Artificial Intelligence Innovations in Cerebrovascular Neurosurgery: A Systematic Review of Cutting-edge Applications

Status:

Version 1

Abstract

Figures

Introduction

Methodology

Results

Discussion

Predicting Clinical Outcomes:

Detection and Diagnosis: Stroke - Swift and Precise:

Enhanced Risk Stratification:

Challenges and Considerations:

AI in Resource-Limited Settings:

Integration into Surgical Workflow

Human Collaboration vs. Machine Dominance

Limitations

Future Prospects

Conclusion

Declarations

References

Tables

Additional Declarations

Status:

Version 1