Spatio-Temporal Features Based Surgical Phase Classification Using CNNs.

Area of Science:

• Medical image analysis
• Computer vision in surgery
• Surgical robotics and automation

Background:

• Accurate surgical phase recognition is crucial for real-time feedback and workflow analysis.
• Laparoscopic cholecystectomy videos contain rich spatio-temporal information for phase classification.
• Existing methods may lack computational efficiency or optimal feature extraction.

Purpose of the Study:

• To develop a novel encoder-decoder based technique for surgical phase classification.
• To leverage spatio-temporal features from laparoscopic cholecystectomy videos.
• To improve classification accuracy and computational efficiency compared to state-of-the-art methods.

Main Methods:

• Utilized an encoder-decoder architecture with the PeleeNet model for feature extraction.
• Employed a combined margin loss function to ensure intra-phase similarity and inter-phase dissimilarity.
• Developed a customized efficient residual factorized CNN (ST-ERFNet) for sequential feature embedding and classification.

Main Results:

• Achieved 86.07% accuracy in classifying 7 surgical phases on the Cholec80 dataset.
• Reduced computational parameters by approximately 84%.
• Demonstrated comparable performance to existing state-of-the-art methods despite reduced complexity.

Conclusions:

• The proposed ST-ERFNet offers an efficient and accurate solution for autonomous surgical phase classification.
• This technology can enhance surgical workflow analysis, error detection, and patient care.
• The method provides a foundation for automated surgical assessment and performance evaluation.