Benchmarking and Enhancing Surgical Phase Recognition Models for Robotic-Assisted Esophagectomy

Yiping Li,Romy van Jaarsveld,Ronald de Jong,Jasper Bongers,Gino Kuiper,Richard van Hillegersberg,Jelle Ruurda,Marcel Breeuwer,Yasmina Al Khalil

2024-12-05

Abstract:Robotic-assisted minimally invasive esophagectomy (RAMIE) is a recognized treatment for esophageal cancer, offering better patient outcomes compared to open surgery and traditional minimally invasive surgery. RAMIE is highly complex, spanning multiple anatomical areas and involving repetitive phases and non-sequential phase transitions. Our goal is to leverage deep learning for surgical phase recognition in RAMIE to provide intraoperative support to surgeons. To achieve this, we have developed a new surgical phase recognition dataset comprising 27 videos. Using this dataset, we conducted a comparative analysis of state-of-the-art surgical phase recognition models. To more effectively capture the temporal dynamics of this complex procedure, we developed a novel deep learning model featuring an encoder-decoder structure with causal hierarchical attention, which demonstrates superior performance compared to existing models.

Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to improve the accuracy of surgical - phase identification in robot - assisted esophagectomy (RAMIE) in order to provide intraoperative support and improve patient prognosis. Specifically, the researchers aim to develop a more effective surgical - phase identification model through deep - learning techniques to meet the challenges brought by the complex multi - anatomical - region operations, repetitive phases and non - sequential phase transitions during RAMIE.

### Main problems:
1. **Complexity**: RAMIE involves multiple anatomical regions, with complex operations, including repetitive surgical phases and non - sequential phase transitions.
2. **Intraoperative support**: It is necessary to provide real - time intraoperative support for surgeons to improve surgical efficiency and safety.
3. **Insufficiency of existing models**: Existing surgical - phase identification models perform poorly in handling complex surgeries such as RAMIE and are unable to effectively capture time - dynamic features.

### Solutions:
- **New data set**: The researchers created a new surgical - phase identification data set, containing 27 videos, specifically for RAMIE.
- **Model comparison**: Several existing state - of - the - art surgical - phase identification models were benchmark - tested using this data set.
- **New model development**: A new model based on the encoder - decoder structure and the causal hierarchical attention mechanism was proposed to better capture the time - dynamic features during the surgical process.

### Specific objectives:
- **Improve identification accuracy**: By improving the model structure and introducing new loss functions, the accuracy of surgical - phase identification is enhanced.
- **Enhance clinical application**: Ensure that the model can be actually applied in the clinical environment to help surgeons perform more precise operations and reduce complications.

Through these efforts, the researchers hope to lay a solid foundation for the development of future surgical - phase identification models and ultimately improve the treatment outcomes of esophageal cancer patients.