Surgical instrument detection and tracking in orthopaedic surgery using computer vision

doi:10.21203/rs.3.rs-4626501/v1

Download PDF

Research Article

Surgical instrument detection and tracking in orthopaedic surgery using computer vision

https://doi.org/10.21203/rs.3.rs-4626501/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

The purpose of this study is to establish a methodology to digitalise the instrument tracking und object tracking analysis in orthopaedic surgery and possibly provide insights about surgical procedures by collecting and utilizing spatiotemporal data about the usage of surgical instruments without introducing new and costly equipment into the operating theatre.

The proposed method utilises a computer vision-based approach. This paper applies an instrument tracking and object detection software on a brief video sequence of an open surgical procedure to generate data that can be utilized for analytical purposes. The object detection module provides categorical and spatial information about each detected instrument, whilst the tracking module assigns a unique ID to facilitate continuous tracking of the instruments detected by the object detection module to keep track of those instruments for the duration of their visibility in the surgical video.

The findings demonstrate that this method successfully generates data for descriptive statistics. These statistics can offer a comprehensive overview of the surgical procedure, contributing to the enhancement of surgical safety and efficiency.

In conclusion the computer vision-based method presented in this study provides an advancement in the digitisation of surgical procedures and object tracking. The ability to collect and analyze spatiotemporal data from standard webcams offers a cost-effective solution, potentially revolutionizing the understanding and improvement of surgical procedures. This method's success in generating descriptive statistics underscores its potential to enhance surgical safety and efficiency, presenting opportunities for further research and application in the field of orthopaedic surgery.

Artificial Intelligence

Computer Vision

Open Surgery

Orthopaedic Surgery

The proficient management and utilisation of instruments in modern surgical procedures constitute a significant challenge that has wide-ranging impacts on both patient safety and operational efficiency. By utilising advanced technologies, the research on surgical instruments stands to undergo a notable transformation through the development of artificial intelligence (AI). Newest AI based software can identify, label and classify data on instrument usage via camera technology. By implementing this technology, healthcare providers can monitor instrument availability and condition in real-time, while systematically minimising errors, optimising operating room workflows, and enhancing patient care. By improving the use and maintenance of instruments, it promotes increased sustainability in healthcare, leading to resources conservation and waste reduction.[1] Incorporating AI systems like these can have a transformative influence on surgical practices.[2]

Surgical training has traditionally focused on the acquisition of skills through observation-based learning and repetitive application and practice. Past experiences and level of expertise of the surgical team are crucial for decision-making during an operation [3]. However, the introduction of surgical data science has the potential to create a paradigm shift in the field by implementing a data-driven approach. Diagnostic and predictive surgical data analytics could lead to better surgical outcomes. This can be obtained through studying potential correlations between data to better understand surgical procedures as well as using already collected data to predict certain outcomes, such as detection of surgical complications [8].

Artificial intelligence (AI) is a rapidly developing field of science and technology that endeavours to equip computers with the objective of equipping computers with the capability to mimic human-like intelligent behaviour. During the second half of the 20th century, there was a notable surge in research activity in this field, which led to the creation of several algorithms, including Support Vector Machines, Bayesian networks, and artificial neural networks. [4, 5]. Several factors have contributed to this advancement, among which are significant improvements in computing power, big data, and increased investment [6]. AI encompasses two facets in the field of medicine: physical and virtual. Physical approaches involve utilizing robotics in surgery to enhance the surgical experience, such as robot-assisted surgery. Among the virtual aspects of the project are software-based solutions, including health management systems and the deployment of innovative guidance systems based on deep learning [7].

However, two general methods are presently employed with the assistance of computers: flowcharts and databases. Flowcharts simulate the approach of medical professionals by taking the patient's medical history and analysing their symptoms in order to make a clinical diagnosis and decision with the greatest possible level of certainty [8]. While the flowchart method has been the traditional approach in simulating medical professional’s decision making, it comes with several inherent limitations. Some of these limitations include that flowcharts can oversimplify complex, multifaceted medical scenarios which cannot be summarised and represented by a flowchart. Further flowcharts lack the ability to learn from experience and adapt based on previous cases, large datasets, and deviations from rules. A flowchart cannot take non-verbal communication and non-pre-defined factors into consideration. To overcome the shortcomings of the flowchart method, the database-based approach is applied, wherein computers can learn from available datasets using machine learning and deep learning models to make clinical decisions. By processing vast amounts of data, AI possesses the potential to uncover previously unknown patterns and associations, augmenting physicians’ knowledge. [4].

Computer Vision is an emerging technology increasingly used in operating theatres to track surgical instruments [4]. Cameras, sensors, and machine learning algorithms are used to capture and analyse images and videos of surgical instruments in real-time and offline. One of the primary applications of computer vision in operating theatres is instrument tracking during procedures [9, 10]. A machine learning algorithm can then provide suggestions for improving surgical techniques.

Object detection and tracking are two important tasks in computer vision. Object detection is a method used to identify and localize objects in images and videos [10], which provides important semantic information. Object tracking, in contrast, fuses an object detection module with a tracking module. The tracking step uses four methods: a feature-, a segmentation-, an estimation-, and a learning-based method. The goal of the tracking step is to perform object identity tracking throughout a sequence of images[11].

Minimally invasive surgery (MIS) has emerged as a favoured surgical approach and standard for surgical video analysis. [12] Nonetheless, there has been a scarcity of investigation and resources to utilise equivalent computer vision approaches in the context of open surgery.

The here presented feasibility study was meant to analyse how deep learning-based computer vision algorithms can be applied in orthopaedic surgery to monitor surgical tools and generate valuable statistics and insights about their usage during surgeries, by means of an object detector and tracker used in conjunction with a basic webcam. A supervised learning approach was chosen, where a machine learning model was trained on a labelled dataset to perform predictions on similar unknown data.

2.1. Dataset

The given dataset included 23636 images of the 14 most used instruments in open orthopaedic surgery. (Table 1) The selection process for these 14 instruments involved carefully choosing a variety of medical instruments that are frequently used in orthopaedic surgeries in order to provide a comprehensive representation of the field.[13, 14] Based on the screening of open source datasets, the medical instruments were classified and examined. The Open-Source Datasets used were HOSPI Tolls Data set [15] and Kaggle.com [16]

The chosen 14 instruments serve as a representative sample of the different surgical instrument groups.

These images were applied for the training of the AI model. The dataset was broadened by self annotated Datasets with 35000 pictures send from an external hospital as well as pictures taken at the Charité University Medicine Berlin together with publicly available open surgery images.[15, 16] Considering that the object detector is based on the concept of supervised learning, the dataset needs to be annotated so that the true labels can be allocated to the objects found in the images. A senior surgeon performed a revision of the dataset to ensure the accuracy of the annotations.

Table 1

Instruments used for the training dataset.
Instruments
1	Artery Forceps
2	Bag and Towel Clips
3	Clamps
4	Diathermy
5	Dissectors and Probes
6	Hollow Ware
7	Needle Holders
8	Retractor
9	Scalpel
10	Scissors
11	Sponge Holding and Sterilizing Forceps
12	Spring Forceps
13	Suction/ Irrigation
14	Tissue Forceps

2.2. Models

Based on our instrument dataset, we trained an object detection model. A deep-learning object-tracking model was combined with an optimal state estimator to provide more accurate tracking of the desired objects. The exact details of these models are the proprietary information of Vision Surgery AI (San Francisco, California, USA) and remain confidential. Figure 1 illustrates the pipeline of our method, where each frame of an input video is initially detected by an instrument detector and then tracked by an object tracker. This figure illustrates our video processing pipeline. Every video frame is processed by the instrument object detector before being forwarded to the tracker to assign an ID for each newly detected object. Throughout the video, the ID serves as the unique identifier for each instrument being tracked. Subsequently, the frame-related information is exported to the output CSV and video files. This process is repeated and saved for each frame.

2.3 Testing

To conduct our feasibility tests, we used a publicly available video that shows a sagittal ligament reconstruction surgery on YouTube [17]. The video shows a sagittal ligament reconstruction by the University of Virginia. We cut the original video and used an 11-second section containing 275 continuous frames. Three types of instruments were utilized: a pair of spring forceps, two Cat Paw retractors, and a pair of Metzenbaum scissors. Processing this video using our proposed pipeline of object detector and tracker, we get an output file in comma-separated values format (csv) showing the type, detection accuracy normalized to a range of 0 to 1, and the location of each instrument detected per frame, in addition to its 4 bounding box identifiers in the Pascal Visual Object Classes (VOC) format. The bounding box identifiers, displayed in the final column of Table 1, represent the top left coordinates (x and y) of the bounding box, followed by the bottom right coordinates (x and y) of the bounding box.

We used the output of the CSV file to measure the distance covered by each tracked instrument in pixels as shown in Fig. 3. The total distance covered by each instrument was calculated by adding up the Euclidean distances between the instrument’s centre across all frames in which it was detected and tracked.

In Table 2, Column 2 lists the instrument IDs assigned by the object tracker to each tracked instrument. The assigned ID number remains the same across consecutive frames as long as the instrument is visible. Figure 2 shows the first and final frame of the utilised video, showcasing the detected instruments along with their corresponding bounding boxes.

Table 2

The output of our combined object detection and tracking model for frames 0 and 1
Frame id	id	Class	Confidence	Bounding Box in VOC format
0	0	Cat paw retractor	0.794	[612, 233, 1281, 482]
0	1	Cat paw retractor	0.853	[283, 1, 432, 126]
0	2	Spring forceps	0.796	[79, 0, 462, 294]
0	3	Metzenbaum scissors	0.649	[329, 253, 496, 560]
∑	0	Cat paw retractor	0.777	[611, 233, 1278, 482]
‚1	1	Cat paw retractor	0.862	[283, 2, 431, 127]
1	2	Spring forceps	0.788	[79, 1, 461, 295]
1	3	Metzenbaum scissors	0.565	[329, 253, 495, 560]

Figure 3 illustrates the distance covered by each tracked tool in pixels. The findings suggest that the two retractors exhibited minimal movement compared to the other three objects being tracked. Based on the graph, it can be inferred that retractors, as shown, demonstrate a comparatively fixed position in contrast to the other tools being studied. The other two instruments depicted being the spring forceps and the Metzenbaum scissors. The graph provides insight into the dynamics of tool movement in surgery. It highlights the differences in change of placement and position among surgical instruments.

This diagram provides a visual representation of the movement dynamics exhibited by each instrument within the video footage. By measuring the distance travelled in pixels, it captures the spatial trajectories, followed by the instruments throughout the duration of the recording. The pixel-based measurement offers a precise quantification of the extent to which each instrument traverses the video frame, allowing for a detailed analysis of its mobility and activity levels. The application of the Euclidean distance metric facilitates a comprehensive understanding of the observed two-dimensional movement patterns in the video dataset. This metric ensures accurate computation of distances between points in the Cartesian plane, facilitating the comparison and interpretation of the tracked instruments’ movements.

Figure 4 represents the spatial distributions of detected instruments in horizontal and vertical image planes as violin plots.

The interquartile range (IQR) is visually represented by black rectangles in these plots, with the median value of the distribution indicated by a white dot within each rectangle. The plot’s width serves as an indicator of the data’s probability density across various levels.

The violin plots show distinct patterns of movement in both horizontal and vertical planes for each instrument. It is worth noting that certain instruments exhibit greater variations in movement compared to others. Retractor_1 and the scissors demonstrate the most extensive distributions of vertical movement, suggesting substantial variability in their motion along the vertical axis. Additionally, the presence of numerous outliers in these distributions indicates occurrences of exceptional movements exceeding the customary range observed for these instruments.

Violin plots provide insights into instrument dynamics, movement patterns, and outliers in surgical scenarios.

The potential challenge in using instrument detection software lies in data privacy, technical reliability and system dependability, and acceptance by medical staff.. To address this issue, this feasibility study explored the use of deep learning-based computer vision algorithms in open surgery, specifically in orthopaedic surgery. It could be shown that the used tracking program was able to identify given surgical instruments from a larger dataset in a video sequence.

The bar graph depicting the total distance covered by each surgical instrument in pixels provides valuable insights into the movement dynamics during the surgical video analysed. The data indicates that the Metzenbaum scissors exhibited the highest degree of movement, followed by the Spring forceps, and the Cat paw retractors. The Cat paw retractor showed the least movement, suggesting it remained relatively stationary during the procedure, which is congruent to the real life application.

These findings highlight the variability in the use and movement of different surgical instruments during an orthopaedic procedure. The substantial movement of the Metzenbaum scissors and Spring forceps mirror the fact that these instruments have an active role in the surgery, such as cutting and holding tissues. In contrast, the relative stability of the Cat paw retractors implies their function as stabilizing tools, which aligns with their intended use to hold tissues apart.

The detailed movement analysis provided by the total distance and violin plots underscores the potential of using computer vision and AI-based tracking systems to enhance the understanding of instrument usage in surgery. This approach can inform training programs by identifying how instruments are used in practice, potentially leading to improved surgical techniques and better patient outcomes.

The current literature mirrors the reduced number of studies testing video instrument tracking. [18–20]

A few studies set up experiments for instrument tracking using electromagnetic tracking [21–24] [25, 26] while only few approaches utilised video tracking. [18–20]

The initial research on the possibility of incorporating spatiotemporal data on surgical instruments relied on either manual reporting of the instruments movements or sensory data collection, such as electromagnetic sensors that capture hand and surgical movement during open surgery[24, 26].

As stated by Genovese et al. [26] the need for more efficient training paradigms have expedited the introduction of video recording and motion analysis systems and virtual reality simulators. In comparison to placing the trackers on the instruments, Datta et al.[24] tried making use of markers located on specific body protuberances to collect information on hand movements. The application of this tracking method would not be feasible in sterile conditions. The sensor would have to be applied to a sterile pair of gloves with a second pair of sterile gloves on top. Sensor placement in this method is prone to a higher level of variability caused by the shifting, slipping, and readjustment of gloves during surgical procedures. Hence, a translation of this tracking method to open orthopaedic surgeries is limited in applicability.

Thus, the study of Genovese et al. [26] points out that instead of focusing on the human hand kinetics which holds higher variability, a focus on tool kinematics offers a standardised procedure with less inter-user variability. The study used electromagnetic sensors placed on the instruments rather than the hands, to track three-axis absolute orientation used for data processing. The findings were dependent on path length, number of movements and time to task completion.[26]

Cavaliere et al. [25] used planar body-mounted electromagnetic sensors to reduce the above mentioned inter-user variability in sensor placement and the resulting measurement. Even with fixed sensors Cavaliere et al. noted that their method proved to be too slow for real time tracking. They noted that they still needed an extensive system amelioration to compensate for metallic distortions from the environment.[25] Here it should be noted that installing body sensor could be difficult to implement in the field of orthopaedic surgeries. Most orthopaedic surgeries use intraoperative x-rays to visualise any fracture reduction or implants used. Thus, the patients and the surgeons have to be adequately protected by wearing lead aprons which might further scew or block the signals send from the body sensory.

While correctly stating some of the strength of electromagnetic tracking, such as small sensor size and no line-of-sight restrictions to take into account Lugez et al. [23] addressed another issue occurring with electromagnetic tracking. The study group reported that electromagnetic measurement accuracy decreased with insertion depth of the surgery as well as tracking at the edges.[23] With videotaping the hurdle of surgery depth and edge tracking could be eliminated. A visual tracking system could be mounted at various angles and zoom settings to ensure a wide scope of camera view.

Because of the high expenses, intricate integration processes of new hardware in the surgical theatre, researchers are now focusing on the use of computer vision algorithms for surgical video analysis [19, 27]. These techniques are remarkably more cost-effective and can operate with as little as a single camera video scoping the surgical field in theatre. By harnessing the power of computer vision models, these methods aim to enhance the safety and comprehensiveness of surgical care. Numerous potential applications of computer vision in surgery have been explored, including surgical skill assessment, surgical phase recognition, and the detection of surgical tools and hands[9] [11]. These investigations have highlighted the versatility and potential of computer vision algorithms in addressing various challenges encountered.

Ganni et al. [20] retrospectively analysed videos of laparoscopically performed cholecystectomies. It was noted that one of the challenges of video tracking is to ensure the instruments’ visibility at all time. Even with numerous cameras and angle optimisation a clear view and the tracking analysis can be impacted if the instrument moves out of the cameras field of view or is covered by tissue or fluids.

One more challenge is the high computational resources needed for training these intricate computer vision models and performing model inference. Training the computer vision models on non-personal data does not need to take place in healthcare facilities and can be performed at institutions with higher computational resources available, such as computational clusters. However, performing predictions in open surgery settings is more critical due to personal data regulations, which allow the processing of the footage on-premises only. To achieve this, high-performance edge devices like Graphic processing units (GPU) are needed, but they come with higher costs and require extra management for resource and data processing.

Certain limitations of the study should be mentioned and taken into consideration. The dataset used in this study entails a video sequence that documents a single surgical procedure, specifically a sagittal band reconstruction. The narrow focus of this dataset may pose limitations when extending the findings to more intricate, complex and lengthy surgical procedures. Complex surgeries may involve a broader array of instruments and a higher degree of variability in surgical techniques. Therefore, the effectiveness of the proposed computer vision-based approach in such settings remains uncertain. Another limitation is the number of instruments in the video. The study results are based on a video sequence from one procedure and the number of instruments utilized in this procedure do not reflect the sheer amount and range of surgical instrument utilized in extensive and highly specialised orthopaedic surgeries. Additionally, lengthy procedures might pose challenges related to camera angles and lighting conditions including shadows and specular reflections. Moreover the preparation necessary for the model to work involving a dataset of previously annotated images can be time-consuming and labour-intensive. The training of the model requires expertise in computer vision as well as surgical knowledge. An annotation inaccuracy can introduce bias and significantly impact the performance and reliability of the model and cause inaccuracy in the results. Lengthy procedures can also pose a challenge for the sheer load of data needing to be processes, stored and analysed after each video tracking session.

The resolution of these hurdles necessitates constant research and collaboration between computer vision specialists and medical professionals to certify the safety and efficacy of computer vision systems in surgical settings.

For the future, it can be postulated that a comprehensive understanding of both computer vision and orthopaedic surgery will be essential for creating multifaceted and useful applications in this field. Therefore, it is imperative to provide education in surgical data science. The goal of this training in data science is to equip healthcare professionals with the ability to incorporate AI into their decision-making processes, leading to enhanced diagnostic accuracy and treatment outcomes. It is equally important for data scientists and engineers to be exposed to clinical issues, allowing them to acquire a deeper understanding of the practical challenges and requirements in a medical context. This interdisciplinary approach fosters the development of more robust and clinically relevant AI solutions.

The here presented program will be further enhanced by integrating supplementary devices and assessing the effectiveness of our approach through multiple orthopaedic procedure videos. By doing so, our aim is to substantiate and improve the precision of our models, ensuring they are reliable and effective in a real-world clinical setting. The ongoing iterative process of model refinement and validation is crucial, as it allows for continuous enhancement of our technique. This approach not only improves the accuracy of AI-assisted surgical insights but also augments decision-making capabilities, ultimately leading to better patient outcomes.

Through iterative testing and feedback, our models can adapt to new data and scenarios, continuously learning and improving. This dynamic process can enhance the accuracy, precision, and efficiency of these tools in the field of orthopaedic surgery. By endorsing collaboration between healthcare professionals and data scientists, we can drive innovation and ensure AI technologies are seamlessly integrated into clinical practice. This integration has the potential to transform the landscape of surgical care and contribute to improved patient outcomes and the advancement of orthopaedic surgery.

Analysing the use of surgical instruments in an operation can be a valuable resource for enhancing surgical practice and decision-making, with a wide range of applications for surgeons, residency training and students. In this feasibility study, a deep learning-based object detector and an object tracker were fused to establish a temporal context and mitigate occlusion effect. In an example video of an open surgery [17] surgical instruments from a given dataset of 14 commonly used instruments could be detected. These outcomes illustrate the potential utility of our technique for enabling advancements in surgical practice and decision-making processes. Our future strategy involves extending our models to encompass a wider range of instruments and evaluating their efficiency in multiple videos of orthopaedic surgery procedures. This expansion aims to create a more comprehensive tool that can be applied across various types of surgeries, thus broadening its applicability.

AI Artificial Intelligence

CSV Comma-separated values

Funding: This research was funded by Vision Surgery AI GmbH.

Institutional Review Board Statement: Not applicable.

Data Availability Statement: The data and models that support the findings of this study are under a non-published license and are not publicly available.

Conflicts of Interest: The authors declare no conflict of interest.

Human Ethics and Consent to Participate declarations: Not applicable.

Author Contribution

Conceptualization, M.P. and T.G.; methodology, M.P..; software, Vision Surgery AI .; validation, T.G., D.B. .; formal analysis, Vision Surgery AI .; resources, Vision Surgery AI; writing M.P., review and editing, T.G. D.B. .;funding acquisition, U.S., T.G All authors have read and agreed to the published version of the manuscript.

Yang, J.H., et al., Using AI and computer vision to analyze technical proficiency in robotic surgery. Surg Endosc, 2023. 37(4): p. 3010-3017.10.1007/s00464-022-09781-y:
Eckhoff, J.A., et al., SAGES consensus recommendations on surgical video data use, structure, and exploration (for research in artificial intelligence, clinical quality improvement, and surgical education). Surg Endosc, 2023. 37(11): p. 8690-8707.10.1007/s00464-023-10288-3:
Maier-Hein, L., et al., Surgical data science for next-generation interventions. Nat Biomed Eng, 2017. 1(9): p. 691-696.10.1038/s41551-017-0132-7:
Gallant, J.N., et al., Patient Perceptions of Audio and Video Recording in the Operating Room. Ann Surg, 2022. 276(6): p. e1057-e1063.10.1097/SLA.0000000000004759:
Bodenstedt, S., et al., Artificial Intelligence-Assisted Surgery: Potential and Challenges. Visc Med, 2020. 36(6): p. 450-455.10.1159/000511351:
Reddy, S., J. Fox, and M.P. Purohit, Artificial intelligence-enabled healthcare delivery. J R Soc Med, 2019. 112(1): p. 22-28.10.1177/0141076818815510:
Bohr, A. and K. Memarzadeh, The rise of artificial intelligence in healthcare applications. Artificial Intelligence in Healthcare, 2020: p. 25-60.
Amisha, et al., Overview of artificial intelligence in medicine. J Family Med Prim Care, 2019. 8(7): p. 2328-2331.10.4103/jfmpc.jfmpc_440_19:
Zhao, Z., et al., Real-time tracking of surgical instruments based on spatio-temporal context and deep learning. Comput Assist Surg (Abingdon), 2019. 24(sup1): p. 20-29.10.1080/24699322.2018.1560097:
Zhao, Z.Q., et al., Object Detection With Deep Learning: A Review. IEEE Trans Neural Netw Learn Syst, 2019. 30(11): p. 3212-3232.10.1109/TNNLS.2018.2876865:
Soleimanitaleb, Z. and M. Keyvanrad, Single Object Tracking: A Survey of Methods, Datasets, and Evaluation Metrics. 2022
Maier-Hein, L., et al., Heidelberg colorectal data set for surgical data science in the sensor operating room. Sci Data, 2021. 8(1): p. 101.10.1038/s41597-021-00882-2:
Andreas Hirner, K.W., Chirurgie: Schnitt für Schnitt. 2. ed. 2008, Stuttgart Georg Thieme Verlag KG.3131308419
Fuller, J.K., Surgical Technology - E-Book: Surgical Technology - E-Book. 2020: Elsevier Health Sciences.9780323680202
HOSPI-Tools Dataset. Last accessed 24/03/2024 ]; Available from: https://zenodo.org/records/6796559).
Kaggle Labeled Surgical Tools and Images Last accessed 24/03/2024 ]; Available from: https://www.kaggle.com/datasets/dilavado/labeled-surgical-tools.
Prof. A. Rashard Dacus, M.D.H., Orthopaedic Hand and Upper Extremity Service,. A. Rashard Dacus, MD, Sagittal Band Hand Reconstruction 2014 [cited 2023; Available from: https://youtu.be/lN7jXyaNWsQ.
Sarikaya, D., J.J. Corso, and K.A. Guru, Detection and Localization of Robotic Tools in Robot-Assisted Surgery Videos Using Deep Neural Networks for Region Proposal and Detection. IEEE Trans Med Imaging, 2017. 36(7): p. 1542-1549.10.1109/TMI.2017.2665671:
Lehr, J., et al., Image-based recognition of surgical instruments by means of convolutional neural networks. Int J Comput Assist Radiol Surg, 2023.10.1007/s11548-023-02885-3:
Ganni, S., et al., A software-based tool for video motion tracking in the surgical skills assessment landscape. Surg Endosc, 2018. 32(6): p. 2994-2999.10.1007/s00464-018-6023-5:
Yaniv, Z., et al., Electromagnetic tracking in the clinical environment. Med Phys, 2009. 36(3): p. 876-92.10.1118/1.3075829:
Schicho, K., et al., Stability of miniature electromagnetic tracking systems. Phys Med Biol, 2005. 50(9): p. 2089-98.10.1088/0031-9155/50/9/011:
Lugez, E., et al., Electromagnetic tracking in surgical and interventional environments: usability study. Int J Comput Assist Radiol Surg, 2015. 10(3): p. 253-62.10.1007/s11548-014-1110-0:
Datta, V., et al., The use of electromagnetic motion tracking analysis to objectively measure open surgical skill in the laboratory-based model. J Am Coll Surg, 2001. 193(5): p. 479-85.10.1016/s1072-7515(01)01041-9:
Cavaliere, M., et al., Planar Body-Mounted Sensors for Electromagnetic Tracking. Sensors (Basel), 2021. 21(8).10.3390/s21082822:
Genovese, B., et al., Surgical Hand Tracking in Open Surgery Using a Versatile Motion Sensing System: Are We There Yet? Am Surg, 2016. 82(10): p. 872-875.10.1177/000313481608201002:
Chadebecq, F., et al., Computer Vision in the Surgical Operating Room. Visc Med, 2020. 36(6): p. 456-462.10.1159/000511934:

No competing interests reported.

Download PDF

Editorial decision: Revision requested
30 Aug, 2024
Reviews received at journal
28 Aug, 2024
Reviews received at journal
26 Aug, 2024
Reviewers agreed at journal
06 Aug, 2024
Reviewers agreed at journal
06 Aug, 2024
Reviews received at journal
27 Jul, 2024
Reviews received at journal
16 Jul, 2024
Reviewers agreed at journal
13 Jul, 2024
Reviewers agreed at journal
13 Jul, 2024
Reviewers invited by journal
13 Jul, 2024
Editor invited by journal
11 Jul, 2024
Editor assigned by journal
09 Jul, 2024
Submission checks completed at journal
09 Jul, 2024
First submitted to journal
23 Jun, 2024

You are reading this latest preprint version

Surgical instrument detection and tracking in orthopaedic surgery using computer vision

Status:

Version 1

Abstract

Figures

1. Introduction

2. Materials and Methods

2.1. Dataset

2.2. Models

2.3 Testing

3. Results

4. Discussion

5. Conclusions

Abbreviations

Declarations

Author Contribution

References

Additional Declarations

Status:

Version 1