[Paper Review] DensiThAI, A Multi-View Deep Learning Framework for Breast Density Estimation using Infrared Images

Siva Teja Kakileti, Geetha Manjunath|arXiv (Cornell University)|Jan 29, 2026

Infrared Thermography in Medicine0 citations

TL;DR

DensiThAI proposes a multi-view deep learning framework to estimate breast density from non-ionizing infrared thermal images, achieving AUROC 0.73 across 10 random splits on a 3,500-woman multi-center dataset using mammography-derived density labels.

ABSTRACT

Breast tissue density is a key biomarker of breast cancer risk and a major factor affecting mammographic sensitivity. However, density assessment currently relies almost exclusively on X-ray mammography, an ionizing imaging modality. This study investigates the feasibility of estimating breast density using artificial intelligence over infrared thermal images, offering a non-ionizing imaging approach. The underlying hypothesis is that fibroglandular and adipose tissues exhibit distinct thermophysical and physiological properties, leading to subtle but spatially coherent temperature variations on the breast surface. In this paper, we propose DensiThAI, a multi-view deep learning framework for breast density classification from thermal images. The framework was evaluated on a multi-center dataset of 3,500 women using mammography-derived density labels as reference. Using five standard thermal views, DensiThAI achieved a mean AUROC of 0.73 across 10 random splits, with statistically significant separation between density classes across all splits (p << 0.05). Consistent performance across age cohorts supports the potential of thermal imaging as a non-ionizing approach for breast density assessment with implications for improved patient experience and workflow optimization.

Motivation & Objective

• Motivate non-ionizing alternatives to X-ray mammography for breast density assessment due to ionizing radiation concerns.
• Investigate whether infrared thermal imaging contains discriminative thermophysical signals corresponding to fibroglandular vs adipose tissue.
• Develop a multi-view deep learning framework to classify breast density from thermal images.
• Validate the framework on a multi-center dataset with mammography-derived density labels as the reference.
• Assess robustness across age cohorts to support potential clinical workflow integration.

Proposed method

• Propose DensiThAI, a multi-view deep learning framework for breast density classification from five standard thermal views.
• Leverage thermophysical differences between fibroglandular and adipose tissues to capture spatially coherent temperature variations on the breast surface.
• Train and evaluate on a multi-center dataset of 3,500 women using density labels derived from mammography as reference.
• Report mean AUROC across 10 random splits with statistical significance testing (p << 0.05).
• Assess performance consistency across different age cohorts to demonstrate robustness.

Experimental results

Research questions

• RQ1 Can infrared thermography reveal discriminative patterns corresponding to breast density that align with mammography-derived labels?
• RQ2 Does a multi-view thermal imaging approach improve breast density classification performance over single-view methods?
• RQ3 Is the proposed DensiThAI framework robust across age groups and multi-center data?
• RQ4 What is the overall diagnostic performance (AUROC) of the framework across multiple random splits?
• RQ5 Are the results statistically significant in separating density classes?

Key findings

• Mean AUROC of 0.73 across 10 random splits.
• Statistically significant separation between density classes across all splits (p << 0.05).
• Evaluation on a multi-center dataset comprising 3,500 women.
• Five standard thermal views used in the framework.
• Consistent performance across age cohorts, supporting non-ionizing infrared imaging viability for density assessment.