AI System Maps the Abdomen to Stage Peritoneal Cancer Without Surgery

Researchers at Dutch medical institutions have developed a deep learning system that can segment the abdomen into 13 distinct regions from standard CT imaging, enabling non-invasive calculation of the Peritoneal Cancer Index (PCI) — a critical metric for staging peritoneal metastases that currently requires exploratory surgery to determine.

The system, published in the International Journal of Computer Assisted Radiology and Surgery, uses architectures including nnU-Net and Swin UNETR alongside the TotalSegmentator framework to achieve automated segmentation of PCI regions. The Peritoneal Cancer Index scores the extent of tumour spread across the abdominal cavity and is used to determine whether patients are candidates for cytoreductive surgery.

Currently, accurately assessing PCI typically requires a surgeon to open the abdomen and visually inspect each region — an invasive procedure that carries its own risks and recovery burden. A reliable imaging-based alternative could spare patients unnecessary surgery, accelerate treatment planning and reduce healthcare costs.

The research represents a practical application of medical AI that addresses a concrete clinical bottleneck. While the system requires further validation in larger patient populations and across multiple imaging centres before clinical adoption, it demonstrates how deep learning can augment surgical oncology by extracting clinically actionable information from imaging data that radiologists find difficult to assess consistently by eye.