Uncovering Latent Pathological Signatures in Pulmonary CT via Cross-Window Knowledge Distillation

2026-05-11

AI Summary

This study addresses the limitation of existing deep learning approaches in multi-window pulmonary CT analysis, which rely solely on late fusion and overlook cross-window pathological interactions. To overcome this, the authors propose a novel cross-window knowledge distillation framework—the first to introduce knowledge distillation into this domain—where a student encoder learns implicit clinical priors from a teacher model trained on the most informative window setting, enabling early fusion of cross-window pathological features. This approach effectively captures latent pathological patterns that are difficult to model with conventional supervision, substantially enhancing generalization. On the COPD-CT-DF dataset, the method improves single-window AUC by 10.1–16.5 percentage points (reaching 0.90–0.94) and achieves an ensemble AUC of 0.9960; it also demonstrates significant performance gains over baselines on the RSNA PE and CTEPH datasets.

Abstract

Multi-window CT imaging captures complementary pathological information across anatomical structures of differing densities, yet existing deep learning methods fuse representations only at later stages, missing cross-density interactions. We propose a cross-window knowledge distillation framework in which student encoders learn latent clinical priors from a teacher trained on the most informative window. Evaluated retrospectively on three cohorts - COPD-CT-DF (n=719), RSNA PE (n=1,433), and an in-house CTEPD dataset (n=161) - distillation improved per-window AUC by 10.1-16.5 percentage points on COPD-CT-DF (0.75-0.81 to 0.90-0.94; all P<0.001), with ensemble AUC reaching 0.9960. Similar gains were observed on RSNA PE (0.80-0.83 to 0.90-0.92) and CTEPD (AUC 0.7481 vs. 0.6264). Cross-window distillation internalises pathological signatures invisible to supervised approaches, offering a generalisable solution for multi-window pulmonary CT analysis.

Problem

• multi-window CT
• cross-density interactions
• latent pathological signatures
• pulmonary CT
• knowledge distillation

Innovation

• cross-window knowledge distillation
• multi-window CT
• latent pathological signatures
• deep learning
• pulmonary imaging