Deep Learning Predicts Imminent Tumor Progression in Advanced Pancreatic Adenocarcinoma Using Serial CT Scans During...

Area of Science:

• Artificial Intelligence in Oncology
• Medical Imaging Analysis
• Deep Learning for Cancer Progression Prediction

Background:

• Advanced pancreatic ductal adenocarcinoma (PDAC) often shows rapid progression during chemotherapy, despite RECIST 1.1 criteria indicating stable disease or partial response.
• Current methods have limitations in accurately predicting short-term progressive disease (PD) in advanced PDAC patients undergoing treatment.
• Timely identification of PD is crucial for modifying chemotherapy regimens and improving patient outcomes.

Purpose of the Study:

• To develop and validate a spatiotemporal deep learning framework for predicting imminent progressive disease (PD) in advanced pancreatic ductal adenocarcinoma (PDAC).
• To dynamically predict PD at the next follow-up visit using serial computed tomography (CT) scans and baseline clinical variables.
• To enable noninvasive, real-time prediction of PD for facilitating timely treatment modifications in advanced PDAC patients.

Main Methods:

• A deep learning framework integrating convolutional neural networks (CNNs) and long short-term memory (LSTM) networks was developed.
• The model was trained on a retrospective cohort of 243 patients, utilizing serial CT scans and baseline clinical data.
• Model performance was evaluated across internal, external, and prospective cohorts, analyzing robustness across different chemotherapy regimens, PD subtypes, and disease stages.

Main Results:

• The deep learning framework achieved significant predictive performance with area under the curve (AUC) values of 0.77 (internal), 0.76 (external), and 0.74 (prospective).
• Robust performance was observed across various chemotherapy regimens (AUC 0.68-0.79), PD subtypes (AUC 0.72-0.77), and baseline disease stages (AUC 0.71-0.85).
• The model demonstrated consistent accuracy in predicting progressive disease, irrespective of treatment type or disease presentation.

Conclusions:

• The developed spatiotemporal deep learning framework provides a reliable method for noninvasive, real-time prediction of imminent progressive disease in advanced PDAC.
• The model's validated generalizability and reliance on routine clinical data suggest its potential for seamless integration into clinical chemotherapy management workflows.
• This AI-driven approach can aid oncologists in making timely treatment decisions, potentially improving outcomes for patients with advanced pancreatic cancer.