Artificial Intelligence in Oncology

Supporting scientific research

Granted applications 2024

Ignace de Hingh

Maastricht University

In short, Ignace de Hingh's research entails the following:

Artificial Intelligence models to improve assessment of peritoneal metastases on imaging

Peritoneal metastases (PM) are predominantly observed in patients with gastric, colorectal and ovarian cancer. Determination of the extensiveness of PM is of paramount importance for assessing treatment eligibility and evaluating treatment response. 

Currently, the gold standard for detecting and quantifying PM is diagnostic laparoscopy (DLS), during which the extensiveness of PM is assessed using the validated Sugarbaker’s Peritoneal Cancer Index (PCI) scoring system. To calculate the PCI, the abdominal cavity is divided into 13 regions, each to which a score between 0 and 3 is assigned based on the size of the largest peritoneal nodule. The final PCI score, ranging from 0 to 39, serves as a good prognostic indicator, is deeply embedded in daily practice, and is used as a cut-off for eligibility for the various treatment options, also within clinical trials.

This project aims to develop a non-invasive method to evaluate the extensiveness of PM, using Artificial Intelligence (AI), to improve the radiological assessment of PM to the level of expert radiologists. Currently, patients undergo imaging, primarily CT scans, for initial staging and treatment response evaluation. However, imaging-based assessment of PM is challenging and requires specific expertise. Applying the PCI scoring system to imaging could provide surgeons and oncologists with an objective and interpretable measure for the extensiveness of PM.

To improve the radiological assessment of PM on imaging, we will develop two AI: an AI model that defines the 13 regions of the PCI on scans, and an AI model that detects and segments PM nodules on scans. Using these AI models, we aim to facilitate standardized and structured assessment of PM on imaging, to reduce interobserver variability among radiologists, and to improve detection performance in PM assessment on imaging.