QS301 - Integrating Artificial Intelligence into Surgical Coaching: A Focus on Retraction Assessment in Total Mesorectal Excision

Tuesday, June 4, 2024

7:51 AM – 7:54 AM ET, USA

Location: Room 327

Quick Shot Presenter(s)

Takuto Yoshida, MD, MPH

Postdoctoral research fellow
University Health Network
University Health Network
Toronto, ON, Canada

All relevant financial disclosures have been managed and will be listed below.
This individual has no financial relationships with ineligible companies.

Author(s)

JH

Jaryd Hunter, BSc

Machine Learning Specialist
University Health Network
Toronto, Ontario, Canada

Disclosure information not submitted.
SP

Sergey Protserov, BSc

Doctoral student
UHN DATA, UHN, Toronto, ON, Canada
Toronto, Ontario, Canada

Disclosure information not submitted.
PM

Pouria Mashouri, BASc, MSc

Manager
University Health Network
Richmond Hill, Ontario, Canada

Disclosure information not submitted.
CM

Caterina Masino, MA, CCRP

Research coordinator
University Health Network
Toronto, Ontario, Canada

Disclosure information not submitted.
Deborah S. Keller, MS MD (she/her/hers)

Associate Professor
Lankenau Medical Center and Lankenau Institute for Medical Research
Lankenau Medical Center and Lankenau Institute for Medical Research
Wynnewood, Pennsylvania, United States

Disclosure information not submitted.
MB

Michael Brudno, PhD

Professor
University Health Network
Toronto, Ontario, Canada

Disclosure information not submitted.
AM

Amin Madani, MD PhD FRCSC FACS

Assistant Professor of Surgery
University of Toronto
Toronto, Ontario, Canada

Disclosure information not submitted.

Senior Author(s)

SC

Sami A. Chadi, MD MSc FRCSC FACS FASCRS

Colorectal surgeon
University of Toronto
Toronto, Ontario, Canada

Disclosure information not submitted.

Purpose/Background: Effective tissue retraction and exposure is critical for optimal surgical performance and patient safety. These critical skills are also not routinely taught during surgical training. With rapid advancements in artificial intelligence (AI), integrating computer vision and deep learning into surgical coaching has great potential for real-time feedback, preoperative case preparation, and postoperative review to improve surgical performance. No AI model to date has assessed retraction during key portions of colorectal surgery. This study aims to develop an AI algorithm to assess the adequacy of tissue retraction during the Total Mesorectal Excision (TME) phase of a low anterior resection, and to evaluate its performance.

Methods/Interventions: An experimental cohort study was performed, using surgical videos of low anterior resections (LAR), specifically examining the medial-to-lateral, proximal posterior, and lateral dissection phases. High-volume colorectal surgeons from 3 countries contributed videos. From each included video, ten random clips were selected, each of which was 5 seconds long and contained 25 frames per second. Videos were annotated by two high-volume surgeons. Tissue retraction was assessed on a 4-point scale: 0- target anatomy not in view, 1- target anatomy in view but with inadequate retraction, 2- target anatomy in view with adequate but suboptimal retraction, and 3- target anatomy in view with optimal retraction (Figure 1). Using this dataset, an AI algorithm was trained and its performance was compared to expert annotations.

Results/Outcomes: A total of 38,368 frames (54 cases) were evaluated. A deep neural network was trained on a dataset split into training (60%), validation (20%), and testing (20%) using the EfficientNet architecture. When comparing inadequate vs adequate/suboptimal and optimal retraction, the model demonstrated an accuracy of 0.78, a specificity of 0.90 and negative predictive value of 0.83. A ROC curve analysis demonstrated the area under the curve (AUC) for the overall dataset and inadequate retraction were 0.76 and 0.75, respectively. Moreover, the model had an inference lag time of 19ms, indicating its ability to be applied real-time during surgery.

Conclusion/Discussion: Adequate retraction to ensure a safe and complete TME is one of the more important and advanced aspects of competency in colorectal surgery. This initial work demonstrates strong potential of AI technology in assessing retraction adequacy during TME. Given the promising results and potential to augment surgical performance and patient safety, future work with a larger sample size and expanded focus in other phases of the colorectal dissection is underway.