Automated Upper Tract Urothelial Carcinoma Tumor Segmentation During Ureteroscopy Using Computer Vision Techniques

Area of Science:

• Urology
• Medical Imaging
• Computer Vision

Background:

• Endoscopic tumor ablation of upper tract urothelial carcinoma (UTUC) offers renal preservation but faces challenges with intraoperative visibility.
• Enhanced visualization is crucial for effective tumor control during UTUC treatment.

Purpose of the Study:

• To develop a computer vision model for real-time, automated segmentation of UTUC tumors.
• To augment visualization during endoscopic UTUC ablation procedures.

Main Methods:

• Collected 20 endoscopic UTUC treatment videos (3387 annotated frames) from two institutions.
• Trained and compared three computer vision models (U-Net, U-Net++, UNext) on annotated frames.
• Evaluated the best-performing model using pixel-based analysis and intraoperative testing.

Main Results:

• The U-Net model achieved high performance in segmenting tumors (AUC-ROC 0.96) and ablated areas (AUC-ROC 0.90).
• A real-time video processing system with intraoperative overlay was implemented.
• The model processed new videos at 15 frames per second.

Conclusions:

• Computer vision models show excellent real-time performance for automated UTUC tumor segmentation during ureteroscopy.
• This technology can significantly improve visualization and aid in renal-sparing treatments for UTUC.