Visualizing the Invisible: Enhancing Radiologist Performance in Breast Mammography via Task-Driven Chromatic Encoding

TL;DR

Dense-breast mammography suffers from masking and high interpretive variability. The authors introduce MammoColor, which integrates a learnable Task-Driven Chromatic Encoding (TDCE) module with BI-RADS triage to produce TDCE-encoded RGB views that align with RGB-pretrained networks and human perception. Across six cohorts and a multi-reader, multi-case (MRMC) study, TDCE improves discrimination (e.g., AUC from $0.7669$ to $0.8461$ on VinDr-Mammo, $P=0.004$) and, in several settings, increases specificity without reducing sensitivity, with pronounced gains in dense breasts and key lesion types. These results support a shift toward human-centered AI visualization that augments radiologists’ perceptual salience and may reduce false positives in screening triage, motivating prospective validation and interface optimization for clinical deployment.

Abstract

Purpose:Mammography screening is less sensitive in dense breasts, where tissue overlap and subtle findings increase perceptual difficulty. We present MammoColor, an end-to-end framework with a Task-Driven Chromatic Encoding (TDCE) module that converts single-channel mammograms into TDCE-encoded views for visual augmentation. Materials and Methods:MammoColor couples a lightweight TDCE module with a BI-RADS triage classifier and was trained end-to-end on VinDr-Mammo. Performance was evaluated on an internal test set, two public datasets (CBIS-DDSM and INBreast), and three external clinical cohorts. We also conducted a multi-reader, multi-case (MRMC) observer study with a washout period, comparing (1) grayscale-only, (2) TDCE-only, and (3) side-by-side grayscale+TDCE. Results:On VinDr-Mammo, MammoColor improved AUC from 0.7669 to 0.8461 (P=0.004). Gains were larger in dense breasts (AUC 0.749 to 0.835). In the MRMC study, TDCE-encoded images improved specificity (0.90 to 0.96; P=0.052) with comparable sensitivity. Conclusion:TDCE provides a task-optimized chromatic representation that may improve perceptual salience and reduce false-positive recalls in mammography triage.

Figures (3)

• Figure 1: Study overview. The workflow comprises three stages: (1) development of a baseline grayscale model and MammoColor (TDCE module + ImageNet-pretrained ResNet-18) after standardized preprocessing; (2) multi-cohort evaluation across public and private datasets; and (3) an external-only MRMC observer study with randomized case allocation to TDCE-encoded versus grayscale reading conditions, stratified by reader experience, using pathology and expert BI-RADS consensus as the reference standard.
• Figure 2: Radar profiles across cohorts (Gray vs TDCE-encoded).
• Figure 3: Qualitative comparison of representative mammographic lesions across grayscale, negative-mode, and TDCE-encoded visualizations. The panels display four major lesion types (from left to right): masses, asymmetries, architectural distortions, and calcifications. For each type, Case (a) and Case (b) illustrate diverse pathological scenarios. The TDCE-encoded views demonstrate enhanced perceptual salience for masses and structural pulling, while grayscale and negative-mode remain complementary for fine calcification analysis.