Gaussian Meta-Space Augmentation for Stacking Ensembles in Multimodal IPMN Risk Stratification

2026-08-11

AI Summary

This study addresses the reliance on invasive biopsies for risk stratification of intraductal papillary mucinous neoplasms (IPMNs) by proposing a non-invasive, ordinal risk prediction method based on multimodal MRI. The approach integrates multi-sequence MRI data, anatomically partitioned radiomic features, and 2.5D CNN-derived representations within a novel stacked ensemble framework termed cUPMI, which employs class-conditional Gaussian augmentation to regularize high-capacity tree-based combiners in a meta-feature space. Evaluated on multicenter data, the method demonstrates significant performance gains: the optimal random forest–based stacked model achieves a quadratic weighted kappa (QWK) of 0.595 and a binary-class AUC of 0.839. Furthermore, cUPMI consistently improves QWK by 0.022 for XGBoost across an 8-stream task, confirming its effectiveness and robustness in ordinal classification with multimodal radiomics.

Abstract

Pancreatic cancer is among the most lethal malignancies; risk stratification of intraductal papillary mucinous neoplasms (IPMNs) offers a crucial opportunity for early intervention but typically requires invasive tissue biopsy. Dominant vision-based approaches, including radiomics and deep learning, provide promising but initially separate discrimination opportunities. Similarly, multisequence MRI (T1W/T2W) and anatomically decomposed (head, body and tail) analysis of the pancreas provide additional and potentially complementary signals. Effective fusion of this information is crucial in ordinal IPMN dysplasia risk prediction and can be accomplished via a meticulously regularized and calibrated ensemble stacking combiner. We present cUPMI, a class-conditional Gaussian augmentation of a combiner's log-probability meta-features, and test it on various prediction paradigms. In our multi-center analysis, we find cUPMI adds limited value to properly regularized L2-logistic binary classification stacks, but consistently regularizes higher-capacity tree combiners in the binary and radiomics-only setting (RF +0.015 and XGBoost +0.024 binary AUC, positive in all seeds). Its cleanest ordinal benefit appears for XGBoost on an 8-stream radiomics task (3-class no < low < high, +0.022 QWK in all seeds). Separately, fold-locked fusion of radiomics and 2.5D CNN streams yields the strongest overall model, an RF stack reaching QWK 0.595 (95% CI [0.54, 0.64]) and binary AUC 0.839, surpassing radiomics, 2.5D ResNet, and 3D DenseNet-121 baselines.

Problem

• IPMN risk stratification
• multimodal fusion
• ensemble stacking
• ordinal prediction
• non-invasive diagnosis

Innovation

• Gaussian meta-space augmentation
• stacking ensembles
• multimodal fusion
• IPMN risk stratification
• class-conditional augmentation