Jiuxu Chen, Thomas J On, Yuan Xu, Jonathan A Tangsrivimol, Kivanc Yangi, Rokuya Tanikawa, Michael T Lawton, Marco Santello, Baoxin Li, Mark C Preul
{"title":"基于人工智能的微血管吻合一致性评估深度学习模型。","authors":"Jiuxu Chen, Thomas J On, Yuan Xu, Jonathan A Tangsrivimol, Kivanc Yangi, Rokuya Tanikawa, Michael T Lawton, Marco Santello, Baoxin Li, Mark C Preul","doi":"10.3171/2025.6.JNS25128","DOIUrl":null,"url":null,"abstract":"<p><strong>Objective: </strong>Assessing the consistency and precision of microanastomosis performance is crucial in neurosurgical training. Traditional methods rely on expert observation, which can be subjective and time-consuming. The aim of this study was to develop and validate a deep learning model using long short-term memory (LSTM) architecture for objective evaluation of microanastomosis performance by predicting and comparing suturing executions.</p><p><strong>Methods: </strong>An LSTM-based neural network was developed to model and predict hand movements during microvascular anastomosis simulation. Video data were collected from 2 expert neurosurgeons performing microanastomosis twice, 1 year apart (sessions 1 and 2). Surgeon 1 performed interrupted suturing, and surgeon 2 performed continuous suturing. Additionally, a trainee with minimal microsurgical experience performed the interrupted suturing procedure once. Model performance was quantitatively assessed by comparing predicted and actual suturing executions using Kullback-Leibler (KL) divergence. Economy and flow of motion were also analyzed.</p><p><strong>Results: </strong>The LSTM-based model accurately predicted suturing movements. Surgeon 1 demonstrated KL divergence values of 0.00063 (session 1) and 0.00061 (session 2), and surgeon 2 had values of 0.00082 (session 1) and 0.00016 (session 2). The trainee exhibited higher KL divergence (0.00196), reflecting less consistent performance. The economy of motion was assessed, showing mean Euclidean distances of 7.41 mm (session 1) and 5.85 mm (session 2) for surgeon 1, 10.53 mm (session 1) and 14.46 mm (session 2) for surgeon 2, and 10.50 mm for the trainee. The flow of motion analysis indicated median time intervals between sutures of 31.96 seconds (session 1) and 29.57 seconds (session 2) for surgeon 1, 21.53 seconds (session 1) and 21.50 seconds (session 2) for surgeon 2, and 101.23 seconds for the trainee.</p><p><strong>Conclusions: </strong>The LSTM-based model objectively assessed microanastomosis performance, capturing consistency and efficiency. Economy and flow of motion metrics were further validated. Future studies will extend the model's application to more surgeons and refine interpretation of the performance metrics.</p>","PeriodicalId":16505,"journal":{"name":"Journal of neurosurgery","volume":" ","pages":"1-10"},"PeriodicalIF":3.6000,"publicationDate":"2025-09-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Artificial intelligence-based deep learning model for evaluating procedural consistency in microvascular anastomosis.\",\"authors\":\"Jiuxu Chen, Thomas J On, Yuan Xu, Jonathan A Tangsrivimol, Kivanc Yangi, Rokuya Tanikawa, Michael T Lawton, Marco Santello, Baoxin Li, Mark C Preul\",\"doi\":\"10.3171/2025.6.JNS25128\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><strong>Objective: </strong>Assessing the consistency and precision of microanastomosis performance is crucial in neurosurgical training. Traditional methods rely on expert observation, which can be subjective and time-consuming. The aim of this study was to develop and validate a deep learning model using long short-term memory (LSTM) architecture for objective evaluation of microanastomosis performance by predicting and comparing suturing executions.</p><p><strong>Methods: </strong>An LSTM-based neural network was developed to model and predict hand movements during microvascular anastomosis simulation. Video data were collected from 2 expert neurosurgeons performing microanastomosis twice, 1 year apart (sessions 1 and 2). Surgeon 1 performed interrupted suturing, and surgeon 2 performed continuous suturing. Additionally, a trainee with minimal microsurgical experience performed the interrupted suturing procedure once. Model performance was quantitatively assessed by comparing predicted and actual suturing executions using Kullback-Leibler (KL) divergence. Economy and flow of motion were also analyzed.</p><p><strong>Results: </strong>The LSTM-based model accurately predicted suturing movements. Surgeon 1 demonstrated KL divergence values of 0.00063 (session 1) and 0.00061 (session 2), and surgeon 2 had values of 0.00082 (session 1) and 0.00016 (session 2). The trainee exhibited higher KL divergence (0.00196), reflecting less consistent performance. The economy of motion was assessed, showing mean Euclidean distances of 7.41 mm (session 1) and 5.85 mm (session 2) for surgeon 1, 10.53 mm (session 1) and 14.46 mm (session 2) for surgeon 2, and 10.50 mm for the trainee. The flow of motion analysis indicated median time intervals between sutures of 31.96 seconds (session 1) and 29.57 seconds (session 2) for surgeon 1, 21.53 seconds (session 1) and 21.50 seconds (session 2) for surgeon 2, and 101.23 seconds for the trainee.</p><p><strong>Conclusions: </strong>The LSTM-based model objectively assessed microanastomosis performance, capturing consistency and efficiency. Economy and flow of motion metrics were further validated. Future studies will extend the model's application to more surgeons and refine interpretation of the performance metrics.</p>\",\"PeriodicalId\":16505,\"journal\":{\"name\":\"Journal of neurosurgery\",\"volume\":\" \",\"pages\":\"1-10\"},\"PeriodicalIF\":3.6000,\"publicationDate\":\"2025-09-26\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of neurosurgery\",\"FirstCategoryId\":\"3\",\"ListUrlMain\":\"https://doi.org/10.3171/2025.6.JNS25128\",\"RegionNum\":2,\"RegionCategory\":\"医学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"CLINICAL NEUROLOGY\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of neurosurgery","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.3171/2025.6.JNS25128","RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"CLINICAL NEUROLOGY","Score":null,"Total":0}
Artificial intelligence-based deep learning model for evaluating procedural consistency in microvascular anastomosis.
Objective: Assessing the consistency and precision of microanastomosis performance is crucial in neurosurgical training. Traditional methods rely on expert observation, which can be subjective and time-consuming. The aim of this study was to develop and validate a deep learning model using long short-term memory (LSTM) architecture for objective evaluation of microanastomosis performance by predicting and comparing suturing executions.
Methods: An LSTM-based neural network was developed to model and predict hand movements during microvascular anastomosis simulation. Video data were collected from 2 expert neurosurgeons performing microanastomosis twice, 1 year apart (sessions 1 and 2). Surgeon 1 performed interrupted suturing, and surgeon 2 performed continuous suturing. Additionally, a trainee with minimal microsurgical experience performed the interrupted suturing procedure once. Model performance was quantitatively assessed by comparing predicted and actual suturing executions using Kullback-Leibler (KL) divergence. Economy and flow of motion were also analyzed.
Results: The LSTM-based model accurately predicted suturing movements. Surgeon 1 demonstrated KL divergence values of 0.00063 (session 1) and 0.00061 (session 2), and surgeon 2 had values of 0.00082 (session 1) and 0.00016 (session 2). The trainee exhibited higher KL divergence (0.00196), reflecting less consistent performance. The economy of motion was assessed, showing mean Euclidean distances of 7.41 mm (session 1) and 5.85 mm (session 2) for surgeon 1, 10.53 mm (session 1) and 14.46 mm (session 2) for surgeon 2, and 10.50 mm for the trainee. The flow of motion analysis indicated median time intervals between sutures of 31.96 seconds (session 1) and 29.57 seconds (session 2) for surgeon 1, 21.53 seconds (session 1) and 21.50 seconds (session 2) for surgeon 2, and 101.23 seconds for the trainee.
Conclusions: The LSTM-based model objectively assessed microanastomosis performance, capturing consistency and efficiency. Economy and flow of motion metrics were further validated. Future studies will extend the model's application to more surgeons and refine interpretation of the performance metrics.
期刊介绍:
The Journal of Neurosurgery, Journal of Neurosurgery: Spine, Journal of Neurosurgery: Pediatrics, and Neurosurgical Focus are devoted to the publication of original works relating primarily to neurosurgery, including studies in clinical neurophysiology, organic neurology, ophthalmology, radiology, pathology, and molecular biology. The Editors and Editorial Boards encourage submission of clinical and laboratory studies. Other manuscripts accepted for review include technical notes on instruments or equipment that are innovative or useful to clinicians and researchers in the field of neuroscience; papers describing unusual cases; manuscripts on historical persons or events related to neurosurgery; and in Neurosurgical Focus, occasional reviews. Letters to the Editor commenting on articles recently published in the Journal of Neurosurgery, Journal of Neurosurgery: Spine, and Journal of Neurosurgery: Pediatrics are welcome.