ORACLE-CT: Anatomy-Aware Support Pooling for CT Classification

2026-06-03

AI Summary

This work addresses the limitation of existing abdominal CT disease classification methods, which rely on global feature aggregation that disregards anatomical structure and thus struggles to align with organ-localized diagnostic evidence. To overcome this, the authors propose the ORACLE-CT framework, which leverages multi-organ segmentation masks to define label-specific anatomical regions of interest and performs attention-based pooling within these regions to enable anatomy-aware feature aggregation. The framework flexibly supports single-organ, multi-organ joint, and local-global strategies, is compatible with various backbones such as DINOv3 and I3D-ResNet-121, and is trained end-to-end. Experiments on MERLIN, Duke-Abdomen, and AMOS datasets demonstrate significant performance gains—for instance, macro-AUROC/AUPRC on MERLIN improves from 0.838/0.638 to 0.858/0.676—while preserving traceability between predictions and anatomical evidence.

Abstract

Abdominal CT disease classification is challenging because each scan is a large 3D volume with many possible findings, while diagnostic evidence is often confined to specific organs or anatomical compartments. Most study-level classifiers aggregate encoder features using anatomy-agnostic pooling or attention, creating a mismatch between localized disease evidence and global evidence aggregation. We propose ORACLE--CT, an encoder-agnostic anatomy-aware aggregation framework that uses multi-organ segmentation to define label-specific anatomical supports and restrict attention pooling to relevant regions. The framework supports single-organ, multi-organ union, comparative, localized, and global support strategies. We evaluate ORACLE--CT with three encoder families: DINOv3, I3D--ResNet-121, and the radiology-native Pillar--0 encoder. Models are trained end-to-end on MERLIN and evaluated internally and under frozen external transfer to Duke--Abdomen and AMOS. Compared with global average pooling, support-masked pooling improved MERLIN macro-AUROC/AUPRC from 0.838/0.638 to 0.858/0.676 for DINOv3 and from 0.829/0.617 to 0.848/0.659 for I3D--ResNet-121. On harmonized 10-label external evaluation, DINOv3 improved on Duke--Abdomen from 0.802/0.628 to 0.835/0.683 and on AMOS from 0.742/0.313 to 0.762/0.350, with similar gains for I3D--ResNet-121. For Pillar--0, most gains came from learned attention, with smaller additional benefit from anatomical masking. ORACLE--CT improves discrimination and external robustness while preserving an auditable link between predictions and anatomical evidence.

Problem

• abdominal CT classification
• anatomy-aware pooling
• localized disease evidence
• global evidence aggregation
• organ-specific support

Innovation

• anatomy-aware pooling
• support-masked attention
• multi-organ segmentation
• CT disease classification
• encoder-agnostic framework