WCLC 2025 - Posters & ePosters-P3.14.16 Prospects of Applyingconvolutional Neural Network Models for of Surgical Videos in Clinical Practice

Pdf Summary

This study from Shanghai General Hospital applied convolutional neural network (CNN) models to improve surgical image analysis for thymic epithelial tumors. Leveraging advances in AI and deep learning, the researchers collected surgical videos from 126 cases—90 simple and 36 complex thymic tumors—and manually annotated key anatomical structures and surgical instruments using LabelMe software. These annotations trained a CNN model capable of performing semantic segmentation to autonomously identify tumor regions and critical anatomical structures such as the phrenic nerve and aorta.<br /><br />The CNN model demonstrated high accuracy, achieving 80%-95% in recognizing surgical instruments and 85% in identifying anatomical features. The study highlights the feasibility of applying semantic segmentation techniques to surgical imaging, which can enhance tumor localization and help preserve vital structures during operations. The authors suggest that this approach may be even more effective in lung cancer surgeries due to the better-defined shape of lung anatomy compared to the thymus.<br /><br />Future refinements using EI VIDEO software aim to further improve the model’s video fitting and predictive performance. This work aligns with the broader trend of incorporating AI technologies like robotic bronchoscopy and 3D reconstruction into thoracic surgery, though practical applications in medical image processing remain limited. The study references related literature in deep learning models for genomics and computer vision analysis of intraoperative videos, showing its foundation in current AI research.<br /><br />In conclusion, this research validates a CNN-based method for semantic segmentation in surgical videos, providing a promising tool for enhancing surgical decision-making and outcomes by accurately identifying tumors and anatomical structures during thoracic surgery.

Asset Subtitle

Shuning Kong

Meta Tag

Speaker Shuning Kong

Topic Mesothelioma, Thymoma, and Other Thoracic Tumors

Keywords

Convolutional Neural Network

Thymic Epithelial Tumors

Surgical Image Analysis

Semantic Segmentation

Deep Learning

Phrenic Nerve Identification

Surgical Instruments Recognition

Thoracic Surgery

AI in Medical Imaging

Robotic Bronchoscopy