Subgroup performance of a commercial digital breast tomosynthesis model for breast cancer detection

2026-03-19 · Source: Machine learning : nature.com subject feeds · Field: Health & Wellbeing — Medical Devices & Health Technology, Clinical Care & Medical Practice, Public Health & Epidemiology · Depth: Advanced, medium

Summary

A study published in Nature Communications on March 19, 2026, presents a detailed subgroup evaluation of a commercial AI model, Lunit INSIGHT DBT v1.1, for breast cancer detection using digital breast tomosynthesis (DBT). Researchers analyzed 167,860 screening exams from female patients, including 1,368 screen-detected cancers, from the Emory Breast Imaging Dataset (EMBED). The model achieved an overall AUROC of 0.91 and a sensitivity of 0.73, demonstrating robust performance across various demographics. However, the study identified lower performance for in-situ cancers (AUROC: 0.85, sensitivity: 0.55), calcifications (AUROC: 0.80, sensitivity: 0.66), and dense breast tissue (AUROC: 0.88, sensitivity: 0.63). Conversely, the model showed higher performance for masses (AUROC: 0.93, sensitivity: 0.85) and architectural distortions (AUROC: 0.90, sensitivity: 0.83).

Key takeaway

For radiologists and AI developers integrating AI into breast cancer screening, you should recognize that while commercial models like Lunit INSIGHT DBT show strong overall performance, they exhibit specific weaknesses. Be vigilant regarding lower detection rates for in-situ cancers, calcifications, and dense breast tissue, and consider these limitations when interpreting results or developing supplementary screening protocols to ensure comprehensive patient care.

Key insights

Granular subgroup analysis reveals a commercial AI model's varied performance in breast cancer detection.

Principles

• AI model evaluation requires detailed subgroup analysis.
• Performance can vary significantly across cancer types and tissue densities.

Method

A retrospective cohort study evaluated a commercial AI model on 167,860 DBT screening exams, stratifying performance by demographic, imaging, and pathologic subgroups to identify disparities.

In practice

• Focus AI development on in-situ cancers and calcifications.
• Consider AI limitations with dense breast tissue.

Topics

• Digital Breast Tomosynthesis
• Breast Cancer Detection
• AI Model Performance
• Subgroup Analysis
• Lunit INSIGHT DBT

Code references

• Emory-HITI/EMBED_Open_Data

Best for: AI Scientist, Research Scientist, Computer Vision Engineer, AI Researcher, AI Data Scientist, Domain Expert

Related on AIssential

• BreastGPT: A Multimodal Large Language Model for the Full Spectrum of Breast Cancer Clinical Routine — Workflow-aligned data and cross-scale visual modeling are critical for clinically grounded medical MLLMs.
• Improving breast cancer screening workflows with machine learning — Google Research, in partnership with NHS organizations, conducted the Artificial Intelligence in Mammography Screening (AIMS) study, publishing two companion studies in Nature Cancer to evaluate an AI…
• Virtual Scanning for NSCLC Histology: Investigating the Discriminatory Power of Synthetic PET — Synthetic PET data can significantly enhance NSCLC histological classification when combined with CT scans.
• Multi-modal AI for comprehensive breast cancer prognostication — A multi-modal AI test significantly improves breast cancer prognostication by integrating pathology and clinical data.
• Parameter-Efficient Adaptation of SAM 3 for Automated ITV Generation from 4DCT Images — Parameter-efficient fine-tuning of SAM 3 with temporal coherence effectively automates ITV generation from 4DCT images using minimal data.
• Preliminary Thoughts On The Midjourney Scanner — Midjourney's new ultrasound tomography scanner faces significant medical limitations and regulatory hurdles for broad diagnostic use.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.