Overview

Schema Version

https://atlas.rsna.org/schemas/2025-11/model.json

Name

Lunit INSIGHT DBT (AI for digital breast tomosynthesis)

Link

https://dx.doi.org/10.1148/ryai.230318

Indexing

Keywords: Breast, Computer-Aided Diagnosis (CAD), Tomosynthesis, Artificial Intelligence, Digital Breast Tomosynthesis, Breast Cancer, Computer-Aided Detection, Screening

Content: BR

RadLex: RID10359, RID45682, RID39055, RID34367, RID10357

SNOMED: 254837009

Author(s)

Eun Kyung Park

SooYoung Kwak

Weonsuk Lee

Joon Suk Choi

Thijs Kooi

Eun-Kyung Kim

Organization(s)

Lunit

Department of Radiology, Yongin Severance Hospital, College of Medicine, Yonsei University

Version

1.0

Contact

moc.liamg@1001krape

Funding

This work was supported by funds secured by Lunit.

Ethical review

Retrospective study approved by ethics review and the central institutional review board; requirement for informed consent was waived.

Date

Updated: 2024-02-28

Published: 2024-04-03

Created: 2023-08-14

References

[1] Park EK, Kwak SY, Lee W, Choi JS, Kooi T, Kim E-K. "Impact of AI for Digital Breast Tomosynthesis on Breast Cancer Detection and Interpretation Time". Radiology: Artificial Intelligence. 2024 May;6(3):e230318. . doi:10.1148/ryai.230318. PMID: 38568095. PMCID: PMC11140510.

Model

Architecture

Deep convolutional neural network (CNN) with a ResNet-34 backbone; trained using neighboring DBT sections around annotated lesions; three-stage pipeline (preprocessing, image analysis, result presentation).

Availability

Proprietary diagnostic support software referred to as Lunit INSIGHT DBT.

Clinical benefit

Improves diagnostic accuracy for breast cancer detection on DBT and reduces radiologist reading time; increases interreader agreement.

Clinical workflow phase

Clinical decision support systems during image interpretation (DBT reading).

Decision threshold

Operating cutoff set on tuning dataset to achieve 90% sensitivity; per-mammogram abnormality score defined as the maximum across the four views used for evaluation.

Degree of automation

Assistive decision support; provides heatmaps and abnormality scores to support radiologist decision-making.

Indications for use

Detection and diagnosis support for breast cancer in women undergoing screening or diagnostic digital breast tomosynthesis (four-view exams; female patients aged ≥22 years) in clinical imaging environments.

Input

Four-view DBT examinations (RCC, RMLO, LCC, LMLO) acquired on Hologic and GE HealthCare systems; with full-field DM or synthetic 2D images available for display alongside 3D sections.

Instructions

The system analyzes input DBT images after cropping/resizing and outputs heatmaps and per-lesion/per-breast abnormality scores; results are presented on DBT slices with a slice navigation bar and corresponding synthetic 2D images within PACS or a dedicated in-app viewer.

Limitations

Reader study used a cancer-enriched dataset; potential laboratory effect; performance differences across lesion types (lower for calcification-only lesions) and vendors; development data contained fewer GE than Hologic cases; system analyzes single examinations without prior studies; limited evaluation for some subgroups.

Output

CDEs: RDE2669, RDE1550

Description: Heatmaps/contour marks localizing suspicious lesions on DBT slices and corresponding synthetic 2D images; per-lesion and per-breast abnormality scores; per-mammogram abnormality score.

Recommendation

Use concurrently with radiologists during DBT interpretation to improve accuracy and efficiency.

Use

Intended: Decision support

User

Intended: Radiologist