Pancreatic cancer is a particularly aggressive type of cancer, with 80 percent of patients who undergo surgery experiencing tumor recurrence within two years. Since tumor tissue and scar tissue can appear very similar on a CT scan, it can be challenging for radiologists to differentiate between the two. Artificial intelligence (AI) can assist in this process by recognizing subtle differences in images, and analyzing large volumes of scans more quickly.

Team Datacation has developed an AI model capable of automatically recognizing and segmenting the pancreas in CT scans. During surgery, both the tumor and part of the pancreas are often removed, resulting in variations in the appearance of the pancreas on postoperative CT scans. Existing segmentation models tend to work suboptimally due to these differences.

The newly developed model utilizes an encoder-decoder architecture with an attention mechanism, allowing it to better adapt to variations among patients. This work has culminated in a scientific publication.

Currently, Datacation is further developing an AI model that can automatically detect tumor recurrence after surgery. To aid radiologists in interpreting the results, significant focus is placed on explainability. For instance, the model can visualize which areas of the CT scan influenced the decision to classify an area as scar tissue or a potential tumor. These visualizations help doctors understand how the model arrives at its conclusions, enabling them to make more confident clinical decisions.

This project is a collaborative effort between UMC Utrecht and Datacation, with co-financing from the European Union. The main objective is to integrate an AI model into clinical practice to assist radiologists in evaluating postoperative CT scans. By enabling quicker and more accurate detection of pancreatic cancer recurrence, treatment can start earlier. This approach promises better treatment outcomes, leading to higher survival rates and an improved quality of life for patients.