Laparoscopic Video Analysis Using Temporal, Attention, and Multi-Feature Fusion Based-Approaches.

Area of Science:

• Medical Artificial Intelligence
• Computer Vision in Surgery
• Surgical Data Science

Background:

• Context-aware systems (CAS) enhance surgical situational awareness and decision support.
• Advances in deep learning have enabled sophisticated analysis of surgical video data.
• Current systems require robust methods for real-time analysis of complex surgical procedures.

Purpose of the Study:

• To develop a deep learning approach for analyzing laparoscopic videos.
• To achieve surgical phase recognition, tool classification, and weakly-supervised tool localization.
• To improve the accuracy and robustness of intelligent systems in the operating room.

Main Methods:

• Adapted ResNet-50 convolutional neural network (CNN) with attention modules and multi-stage feature fusion.
• Utilized a multi-map convolutional layer with pooling for tool localization and presence detection.
• Employed a long short-term memory (LSTM) network for temporal modeling, tool classification, and phase recognition.

Main Results:

• Achieved 88.5% precision and 89.0% recall for surgical phase recognition.
• Reached 95.6% mean average precision for tool presence detection.
• Obtained a 70.1% F1-score for tool localization, demonstrating effective feature learning.

Conclusions:

• The proposed deep learning model effectively analyzes laparoscopic videos for critical surgical tasks.
• Integrating attention modules and multi-stage feature fusion enhances the robustness and precision of surgical analysis.
• This approach shows significant potential for advancing intelligent context-aware systems in future operating rooms.