Combining AI-based image biomarkers to predict pancreatic ductal adenocarcinoma survival
Keywords
Loading...
Authors
Issue Date
2025-03-18
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
Pancreatic ductal adenocarcinoma (PDAC) has the lowest 5-year survival rate of all cancer
types and is accompanied by challenges at every step of the patient’s pathway. There has
been an increase in AI research to advance cancer care, however, as PDAC has a lower inci
dence than other forms of cancer, it is not as well-researched and there is an absence of proven
prognostic biomarkers for PDAC. This project aimed at investigating the prognostic value of
combining AI-based image biomarkers from histopathological images of patients with PDAC
when predicting overall survival. Several algorithms were used to quantify these biomarkers
from whole-slide images (WSIs) and they have previously been found to be individual prog
nostic factors. As these algorithms required segmentation masks of several tissue types, this
project attempted to improve the segmentation performance of an existing multi-tissue seg
mentation baseline by retraining this baseline. The retrained multi-tissue segmentation model
achieved a similar performance with less variance when using a smaller and more efficiently
trained model. In addition, the retrained model showed better generalization than the baseline
on unseen data. The resulting masks were then given to the biomarker extraction algorithms
to attempt inferring informative features. These biomarkers could finally be used, with a Cox
regression and a logistic regression model, for survival prediction. The findings indicated that
using a combination of AI-based image biomarkers does not outperform logistic regression
models trained on one of these biomarkers for predicting PDAC survival. Our survival analysis
did show promising potential for using mitosis density as a potential prognostic biomarker for
pancreatic cancer with statistical significance.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen