Lunit INSIGHT MMG

Lunit INSIGHT MMG is an artificial intelligence-powered software designed to enhance breast cancer detection by aiding physicians in the analysis of screening mammograms. Utilizing a deep learning algorithm trained on diverse radiological data, it identifies suspicious lesions and provides quantitative abnormality scores as an adjunctive tool, demonstrably improving diagnostic accuracy.

Lunit INSIGHT MMG is a radiological Computer-Assisted Detection and Diagnosis (CADe/x) software designed to assist physicians in interpreting screening mammograms for breast cancer. It functions as an adjunctive tool, intended to be viewed after the physician's initial read, to aid in the detection, localization, and characterization of suspicious areas on images from compatible Full-Field Digital Mammography (FFDM) systems. The core technology relies on an artificial intelligence algorithm, specifically a deep learning model trained on extensive databases of biopsy-proven breast cancer cases, benign lesions, and normal tissues. The software automatically analyzes de-identified DICOM mammogram images, identifies suspicious lesions, and provides a visualized map along with an abnormality score for each lesion and per breast, reflecting the likelihood of malignancy. The device's performance was evaluated through standalone testing and a multi-reader multi-case (MRMC) clinical study. Standalone performance tests on 2412 mammograms demonstrated a ROC AUC of 0.903, with a sensitivity of 85.74% and specificity of 75.62%, indicating improvement compared to unaided radiologist interpretation. The MRMC reader study, involving 240 mammograms and 12 radiologists, showed that using Lunit INSIGHT MMG significantly enhanced radiologists' diagnostic ability; the average ROC AUC increased from 0.754 (unaided) to 0.805 (aided). Secondary endpoints, including sensitivity, also improved with CAD assistance, and the software’s standalone performance on the reader study dataset (ROC AUC 0.863) surpassed that of unaided radiologists, suggesting a potential clinical benefit in reducing false positives and false negatives.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the 'abnormality score' generated by Lunit INSIGHT MMG correlate with BI-RADS categories, and what is the distribution of these scores for malignant versus benign findings across diverse patient subgroups?; What were the specific characteristics of the 'large databases of biopsy proven examples' used to train the deep learning algorithm, particularly regarding the diversity of lesion types, breast densities, and patient demographics?; Given that the Lunit INSIGHT MMG is intended for use after the initial read, what was the observed impact on the overall reading time per case in the clinical study, and how does this compare to concurrent reading CAD systems?; How does the performance of Lunit INSIGHT MMG, specifically its sensitivity and specificity, vary when applied to mammograms from different FFDM equipment manufacturers (e.g., Hologic, GE, Siemens) and across the full spectrum of breast density categories?; The standalone performance ROC AUC (0.903) was notably higher than the aided-reader ROC AUC (0.805). What factors might contribute to this difference, and what strategies could optimize human-AI interaction to better translate the AI's standalone capabilities into improved clinical outcomes?