JDCNet: Confidence-Gated Privileged-Modality Distillation for Cost-Preserving X-ray Inference

TL;DR

This paper introduces JDCNet, a distillation framework that uses privileged CT modality during training to improve X-ray inference efficiency, matching the deployment cost of a single X-ray modality.

Contribution

JDCNet employs confidence-gated distillation from CT to X-ray, demonstrating improved performance over supervised baselines in a large patient cohort.

Findings

JDCNet outperforms supervised ResNet-18 baseline in BA by +0.035 and +0.033 with different supervision methods.

Confidence-gated auxiliary targets are more effective than uniform softening of CT logits.

Other distillation and transfer methods tested did not surpass the proposed confidence-gated approach.

Abstract

We study a systems-level visual inference problem: using an expensive privileged modality during training while preserving a fixed-cost, single-modality deployment path. We present JDCNet, a confidence-gated CT-to-X-ray distillation framework in which the CT teacher supplies an auxiliary hard or temperature-scaled target only on training samples whose teacher confidence exceeds a threshold; at deployment the student takes X-ray input alone and matches the parameter, MAC, and latency profile of the supervised X-ray baseline. On a 510-patient same-patient paired BIMCV cohort with patient-level 5-fold cross-validation, two JDCNet configurations clear a fixed transfer gate against the supervised ResNet-18 baseline: 3-slice soft-KL supervision yields $\Delta\mathrm{BA}{=}{+}0.035$ ($95\%$ CI $[{+}0.011,{+}0.057]$) and mid-slice hard supervision yields $+0.033$ ($[{+}0.007,{+}0.058]$). Under the same splits and gate, logit distillation, gated logit distillation, contrastive alignment, attention transfer, feature hints, BiomedCLIP fine-tuning, and a module-augmented variant do not pass. Confidence-gated auxiliary targets are therefore a more transferable channel than uniformly softened CT logits; the evidence is bounded to one paired cohort, so external paired-cohort replication is required before any deployment claim.