Radiomics Boosts Deep Learning Model for IPMN Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Intraductal Papillary Mucinous Neoplasm (IPMN) cysts are pre-malignant pancreatic lesions with the potential to develop into pancreatic cancer.
• Accurate risk stratification of IPMN is critical for effective treatment planning and disease management.
• Challenges in IPMN detection arise from the complex morphology of cysts and the pancreas, often requiring advanced imaging analysis.

Purpose of the Study:

• To develop and validate a novel computer-aided diagnosis (CAD) pipeline for risk classification of IPMN cysts.
• To enhance the accuracy of IPMN risk stratification from multi-contrast MRI scans.
• To improve clinical decision-making for patients with pancreatic cystic lesions.

Main Methods:

• A novel computer-aided diagnosis pipeline integrating volumetric self-adapting segmentation and a deep learning classification scheme.
• A radiomics-based predictive approach combined with deep learning for enhanced classification accuracy.
• Validation on multi-center datasets comprising 246 multi-contrast MRI scans from five institutions.

Main Results:

• The proposed decision-fusion model achieved superior performance compared to the state-of-the-art in IPMN risk classification.
• An accuracy of 81.9% was attained, significantly outperforming existing international guidelines and published studies (61.3%).
• Ablation studies confirmed the critical contributions of both radiomics and deep learning modules to the model's performance.

Conclusions:

• The developed AI pipeline demonstrates high accuracy and robustness for IPMN risk classification from MRI scans.
• This advanced tool has significant implications for improving clinical decision-making and patient management for IPMN.
• The findings represent a substantial advancement in the non-invasive stratification of pre-malignant pancreatic lesions.