Clinical Features And Neuroimaging Findings In Acute Cerebral Infarction Patients Using RAPID Articial Intelligent (RAPID AI) Software Analysis – A Series Of 54 Cases

Background: Stroke is the second leading cause of death and the leading cause of permanent disability globally. Vietnam is a developing country with a high prevalence of stroke but is under-resourced in terms of specialist staff able to interpret complex brain imaging. Methods: A case series of 54 stroke patients admitted between October 2019 and October 2020 where thrombectomy was being considered and where ‘Rapid’ Articial Intelligence (AI) was used to analyze images of stroke. Results: The mean age of patients was 73.39 ± 12.46 years with 57% male. The most common risk factors were (76%), atrial brillation (24%), diabetes (20%), alcohol (15%), and smoking (9%). The most common clinical signs were hemiparesis in 76% of the patients, followed by dysphasia in 50% and memory loss in 28% of the sample. 7% presented with dizziness and 7% with headache. 6% were unconscious on admission. ASPECTS evaluation showed that 24 (44%) patients had good ASPECTS scores of 8-10, 17 (32%) patients had ASPECTS scores of 5-7, and 13 (24%) patients had ASPECTS scores of 0-4. The number of patients with an infarct core volume <70 mL was 50 (93%), while a mismatch volume of >15 mL was observed in 31 (55%) patients and 22 (41%) patients had a mismatch ratio >1.8. The assessment of CT imaging of thrombi showed 51 cases of anterior cerebral circulation, including 13 (24%) cases diagnosed as ICA, 30 (76%) cases diagnosed as MCA, and 8 (15%) cases diagnosed as SA. There were 10 cases of MCA-M1 (19%), 7 cases of MCA-M2 and MCA-M4 (13%) cases and 6 cases of MCA-M3 (11%), respectively. There were three cases of posterior cerebral circulation, comprising Posterior Cerebral artery (PCA) infarction and two cases of Basilar Artery (BA) territory infarction. Conclusions: RAPID Articial Intelligent (RAPID AI) Software Analysis combined with clinical assessment can be used in identify the size and site of cerebral infarction and diffusion perfusion mismatch in


Background
Stroke is the second leading cause of death and the leading cause of disability globally [1]. In recent years, the burden of stroke associated with mortality, morbidity, and disability has been increasing across the world [2,3]. Risk factors associated with stroke can be divided into groups of modi able and nonmodi able features. The group of modi able risk factors includes hypertension, smoking, diet, and physical inactivity, while the group of nonmodi able risk factors comprises age, sex, family history, and race/ethnicity [4].
Vietnam is a developing country with nearly 100,000 million people, and strokes and stroke treatment are a major health burden in Vietnam [2,5,6]. About 200,000 people have a stroke each year with a 50% mortality rate, and with only 10% of patients returning to a normal life after stroke [6]. The facilities for the diagnosis and treatment of stroke are limited. The diagnosis of stroke is currently based on clinical ndings combined with computer tomography imaging (CT) or magnetic resonance (MRI), however CT imaging or MRI facilities are not readily available and affordable in all hospitals and access to neuroradiology expertise is very limited.
Phu Tho Provincial General Hospital (GHP) is the leading provincial public hospital and the largest hospital in the northwest of Vietnam. The hospital has 1500 beds and well-equipped facilities for image diagnosis with both modern CT and MRI machines. The stroke center of GHP was established in 2018, containing 150 beds for inpatient treatment of people who suffered acute stroke and who require rehabilitation after stroke. In June 2019, the hospital began to use RAPID arti cial intelligent (RAPID AI) software (iSchemaView, USA), for the imaging analysis of stroke characteristics based on an arti cial intelligence technique.
This study presents the diagnostic and treatment features of stroke patients based on RAPID AI software assessment in Phu Tho Provincial General Hospital in Vietnam in 2019-2020.

Study Design
This study was a case series based on the data from electronic medical records of stroke patients admitted to Phu Tho Provincial General Hospital between October 2019 and October 2020, for whom Rapid AI was used to analyze images of stroke. The study was testing the practicability and value of using the technology in routine practice in Vietnam. Patients The stroke patients studied in this study had Rapid AI used for the diagnosis and analysis of cerebral images. All patients had CT or MRI scans performed within 24-48 h after the onset of stroke and were being considered for late thrombectomy. Information on the patients was collected using electronic medical records (Fig. 1). All patients gave written informed consent for participation in the study. Demographic Characteristics Demographic data collected included age, sex, work status, living area, time onset of stroke, reasons for admission, hospitalization status, treatment duration, and discharge status.

Risk Factors for Stroke and Clinical Presentation
Information concerning risk factors for stroke, including age, sex, hypertension, diabetes, atrial brillation, smoking, and alcohol habits, was extracted from the electronic medical record pro les. The smoking habits, alcohol consumption, and onset time were determined in the present study by a neurologist questioning patients and/or relatives who had observed the onset and who were aware of the patient's recent health and clinical characteristics on admission were recorded Laboratory Investigations Diabetes was de ned as a fasting blood glucose of greater than 7mmol/l or use of hypoglycemic medications, and hyperlipidemia as serum levels of total cholesterol of greater than 5.2 mmol/dl. Sociodemographic characteristics CT Imaging and Perfusion Imaging Analyzed Based using RAPID AI On admission, ischemic stroke was diagnosed using clinical ndings and brain imaging. The CT perfusion image characteristics were characterized in terms of ASPECTS, infarct core volume, and mismatch volume. Stroke patients were divided into four categories as follows: total anterior circulation infarcts, partial anterior circulation infarcts, posterior circulation infarcts, and lacunar infarcts. The imaging data of all patients were analyzed independently by two experienced neurologists. Lesions were divided into the following four categories: infarction of the anterior cerebral artery territory, middle cerebral artery territory, posterior cerebral artery territory, and vertebrobasilar artery territory. Cerebral imaging parameters were used to determine the locations of the lesions and then to identify infarctions in either a single vascular territory or multi-vascular territories (vascular territories ≥ 2). Statistical Analysis SPSS (version 22.0; IBM Corp., Armonk, NY, USA) was used for data analysis. Descriptive statistics were used to summarize sociodemographic and clinical information as well as subclinical data. Categorical variables were expressed as frequencies and percentages. Ethics Approval The study was approved by the Ethics Committee of Phu Tho Provincial General Hospital. The con dentiality of the information regarding patients was ensured in such a way that the data could only be used for the study purpose. The information obtained from the patients' electronic medical records is also presented only in a collective manner.

Patient Characteristics
In total, 54 electronic medical records of patients diagnosed with ischemic stroke were identi ed. The age range of the 54 ischemic stroke patients was from 18 to 95 years of age. The mean age of the patients was 73.39 ± 12.46 years old, and the majority of patients (85.2%) were in the age group of > 60 years old. Male patients comprised 31 (57.4%) of the total patients. The majority of the patients were retired (42.6%) and living in rural areas (64.8%). The time from onset to hospital arrival ranged from 0-6 h for 28 patients (51.9%), from 6 to 24 h for 24 patients (44.4%), and beyond 24 h for 2 patients (3.7%). Patients admitted to hospital were mostly in states of unconsciousness (50.0%) and coma (7.4%), while 42.6% of patients were conscious. The number of patients discharged from the hospital who had made some recovery was 40 (74.0%), while 7 (13%) patients became worse or died ( Table 2).

CT Perfusion Images Based on RAPID AI Analysis
The distribution of the ASPECTS score is given in Table 5. Only 4 (7.4%) patients had an infarct core volume greater than 70 mL compared to 50 (92.6%) patients with an infarct core volume smaller than 70 mL. The mismatch volume was greater than 15 mL in 31 (54.7%) patients, while 23 (45.3%) patients had a mismatch smaller than 15 mL. In total, 32 (59.3%) patients had a mismatch ratio under 1.8, but 22 (40.7%) patients had a mismatch ratio over 1.8 (Table 5). In our study, we observed 51 cases of anterior cerebral circulation including 13 (24.07%) cases that were due to ICA occlusion, 30 (75.93%) cases diagnosed as MCA occlusion, and 8 (14.81%) cases diagnosed as Small Artery occlusion (SA). Of these, the number of cases of MCA-M1 was 10 (18.5%), the number of cases of each MCA-M2 and MCA-M4 was 7 (13%), and the number of cases of MCA-M3 was 6 (11%), respectively. There were only three cases of posterior cerebral circulation, including one (1.9%) case of PCA and two (3.7%) cases of BA. In total, 28 stroke patients were admitted to the hospital within the rst 6 h after onset, and 24 patients were admitted within 6 to 24 h. Only two patients were admitted beyond 24 h (Table 6). Table 6. Cerebral occlusive position based on the RAPID AI analysis (n = 54).

Discussion
Phu Tho Provincial General Hospital was the rst hospital in to Vietnam introduce RAPID AI technology for the diagnosis and treatment of stroke patientsThis study presents the results of the diagnosis and treatment of stroke patients for whom RAPID AI combined with neuroimaging was used for the detection, characterization, and prognostication of acute strokes. AI technology is a rapidly developing eld and represents a promising avenue for fast and e cient imaging analysis [7]. RAPID AI has been approved and certi ed by the FDA for perfusion imaging and is currently used in 50 countries. RAPID AI software can be used for perfusion imaging in stroke. The RAPID AI software was validated in the DEFUSE 2 study in 2012 and received FDA approval in 2013 [8]. RAPID software analyzes CT and MRI perfusion within 3 minutes and generates colorimetric perfusion maps of stroke penumbra as well as the core mismatch volume and mismatch ratio [9]. The penumbra mismatch sensitivity is 100% and speci city is 91% [10][11][12][13]. RAPID AI technology was used in the recent large LVO ET trials EXTEND IA, SWIFT PRIME, CRISP, DEFUSE 2 and 3, and DAWN [8, 10, 14-16].
Stroke incidence increases with age and is more common in males [17]. The time from symptom onset to hospital arrival (time to hospital) is a key factor for delivering effective treatment and improved outcomes of stroke patients, especially for patients with ischemic stroke. Our results for time to hospital showed that about half of patients arrive after six hours and are therefore less likely to be eligible for reperfusion treatment. If outcomes are to improve in Vietnam the systems for prehospital care in terms of patient and population awareness of the symptoms of stroke and emergency medical services need to improve. The patients who do present are predominantly very severe stroke with over half having reduced levels of consciousness. These results were similar to those of a study in Hong Kong [22] and Shanghai [23], which showed that 56.3% and 51.9% of patients arrived at a hospital within 6 h after the rst symptom of stroke. Previous studies have presented a median arrival time that varied from 2.51 to 15 h [24]. However, studies on the time to hospital of patients in European countries reported that the majority of patients arrived within 3 h of onset, earlier than in Asian countries, which is due to the higher rate of ambulance transport of stroke patients and better stroke awareness knowledge in these countries [25].
In our study, we found that the most common risk factor was hypertension, identi ed in 75.9% of patients, followed by atrial brillation in 24.1% of patients, and diabetes in 20.4% of patients. These results align with previous studies indicating that uncontrolled hypertension is the most important risk factor for stroke in developing and developed countries [26]. This may re ect the fact that hypertension has been identi ed as the most prevalent modi able risk factor and is a powerful modi able risk factor. On the other hand, the numbers of stroke patients with smoking and alcohol habits in this study were lower (< 15%) than in other studies, [27].
Hemiparesis and dysarthria were two of the most common clinical presentations of stroke in our study, at 75.9% and 50% of the total, respectively. This is similar to cohorts reported from Egypt [28] and India [29], which showed that 76.1% of patients had hemiparesis and 60% of patients had dysarthria. Memory loss commonly occurs as a result of stroke and was observed in 27.8% of patients in our study. Other investigators have reported a dementia incidence of approximately 25% at 3 months after ischemic stroke. In the current study, a small number of patients suffered from dizziness and headache (7.4%), and unconsciousness and near death (5.6%). These results also agree with the results of El Tallawy [28].
There was a variety in the laboratory tests requested as this depends on the stroke clinicians rather than a formal protocol In our study, the laboratory tests included BP, RBC, BG, cholesterol, PC, INR, and triglyceride. These tests were examined to evaluate patients before the administration of tPA. Some of the tests are necessary to determine suitability for intravenous thrombolysis but some are often unnecessary and just delay commencement of acute treatments. [32].
The application of RAPID AI to read CT perfusion images was rst applied in Phu Tho Provincial General Hospital, Vietnam, in June 2019. RAPID AI technology processing takes within 3 minutes to send the results of ASPECTS, infarct core volume, mismatch volume, and mismatch ratio to PACS and RAPID mobile apps. This application was developed by an experienced neurologist, Greg Alber [33]. The ASPECTS is a 10-point quantitative topographic CT scan score used for patients with middle cerebral artery (MCA) stroke [34]. Our study showed the ASPECTS score was good in 44.4% of patients, compared with 31.5% showing a bad score and 24.1% showing the worst possible score. The assessment of the ASPECTS score has been used to direct therapies. Those with a low ASPECTS score, suggesting a large MCA infarction, can be excluded from futile intra-arterial treatments, which are unlikely to result in patients gaining functional independence and increase the risk of hemorrhage [34]. The infarct core volume shows the part of the AIS which was already infarcted or was irrevocably destined to infarct regardless of reperfusion [35]. The inclusion criteria used to select patients eligible for interventional thrombectomy in the case of patients admitted to hospital within or beyond 6 h onset include an initial infarct volume of < 70 mL, a ratio of ischemic penumbra to infarct core of ≥  [11,38]. The evidence of using RAPID showed the diffusion-perfusion mismatch identi ed by RAPID was in agreement with the observation of a human reader in 60 cases (95.2%) and in disagreement in 3 cases (4.8%, 3 false positives) [11]. RAPID was able to identify mismatches with 100% sensitivity and 91% speci city (false positive rate = 9.1%, false negative rate = 0%) in analyses with the observations of the human reader as a ground truth [11].
The RAPID software enabled information about the location, size, and whether there was any potentially salvageable brain enabling the physicians to make decisions on the suitable treatment, whether thrombolysis or thrombectomy [39]. The diagnostic evaluation of occlusive thrombi on noninvasive studies now constitutes an integral component of acute stroke management [39]. In a hospital with limited access to neuroradiology 24 hours a day, seven days a week, such a facility is of critical importance and has huge potential to improve the quality of clinical care in hospitals and countries with limited resources. Clearly, this needs to be combined with improving public awareness that stroke is a treatable condition if they get to hospital quickly after the onset of symptoms, and an improved emergency medical service. The current poor outcomes with a 50% mortality and only 10% making a full recovery are unacceptable. Using modern technology such as RAPID or similar arti cial intelligence systems will be important to improve outcomes.

Conclusions
The majority of studied patients in this study had severe stroke and presented late. While work is needed to improve prehospital care the value of arti cial intelligence software to identify which patients might still bene t from reperfusion will be important, particularly in resource poor settings to improve the quality of care and outcomes. This is the rst study to present results on the diagnostic and treatment features of stroke patients in Vietnam for whom RAPID AI was used for their screening and management. Table 2   Table 2 is not available with this version Figures Figure 1 Flow diagram of the study and selected eligible patients based on RAPID