Impact of neoadjuvant chemotherapy on breast tissue density and its correlation with pathologic response: a retrospective AI-Enhanced radiological study.
Authors
Affiliations (13)
Affiliations (13)
- Università degli Studi di Milano, Via Festa del Perdono, 7, 20122 Milano MI, Italy; Breast Imaging Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
- Postgraduation School in Radiodiagnostics, Università degli Studi di Milano, Milan, Italy. E-mail: [email protected].
- Department of Medical and Surgical Sciences, Section of Diagnostic Imaging, University of Foggia, Foggia, Italy, E-mail: [email protected].
- Postgraduation School in Radiodiagnostics, Università degli Studi di Milano, Milan, Italy. E-mail: [email protected].
- Breast Imaging Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
- Breast Imaging Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
- Breast Imaging Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
- Diagnostic and Interventional Radiology Department, IRCCS Cà Granda Fondazione Ospedale Maggiore Policlinico, Università degli Studi di Milano, Milan, Italy, E-mail: [email protected].
- Breast Imaging Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
- Breast Surgery Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
- Department of Medical and Surgical Sciences, Section of Diagnostic Imaging, University of Foggia, Foggia, Italy, E-mail: [email protected].
- Diagnostic and Interventional Radiology Department, IRCCS Cà Granda Fondazione Ospedale Maggiore Policlinico, Università degli Studi di Milano, Milan, Italy, [email protected].
- Breast Imaging Division, IEO European Institute of Oncology IRCCS, Milan, Italy, E-mail: [email protected].
Abstract
To evaluate changes in mammographic breast density after neoadjuvant chemotherapy (NAT), their association with pathologic response, and agreement in density classification among human readers and an artificial intelligence (AI) tool. This retrospective study included 135 women with triple-negative and/or HER2-positive invasive ductal breast cancer who underwent NAT and paired pre- and post-treatment digital mammography. Breast density was assessed by three radiologists, one non-radiologist reader, and an in-house AI tool. The AI model was developed on an independent dataset of 10,000 mammograms, with manual fibroglandular masks available for a stratified random subset of 2,000 examinations. Paired density changes were analyzed using the Wilcoxon signed-rank test and Stuart-Maxwell test; predictors of density reduction with ordinal logistic regression; pathologic response correlation with Spearman's rho; and reader agreement with quadratic-weighted kappa and intraclass correlation coefficients (ICC). The paired four-category density distribution changed significantly after NAT (Stuart-Maxwell chi-square = 35.19, df = 3, p < 0.001). ACR C decreased from 31.9% to 17.8% and ACR D from 14.8% to 6.7%; adjusted exact McNemar p values were 0.001 and 0.013. Greater density reduction was associated with younger age, premenopausal status, and higher baseline density. Density reduction correlated with pathologic response (Spearman rho = 0.53; 95% CI, 0.40-0.64; p = 3.8 x 10^-11). Dice coefficients were 0.88, 0.87, and 0.86 for training, validation, and test sets. Expert-AI agreement was good (ICC = 0.78; 95% CI, 0.70-0.85). NAT was associated with reduced mammographic density, and greater reductions were associated with more favorable pathologic response. The AI tool showed good agreement with the expert reader. This retrospective single-centre study suggests that mammographic density tends to decrease after NAT and that AI-assisted density classification can improve assessment consistency. These findings are exploratory and support prospective external validation rather than immediate biomarker adoption.