Digital mammography with AI-based computer-aided diagnosis to predict neoadjuvant chemotherapy response in...

To investigate whether digital mammography (DM) with artificial intelligence-based computer-aided diagnosis (AI-CAD) predicts pathologic complete response (pCR) after neoadjuvant chemotherapy (NAC) in human epidermal growth factor receptor 2 (HER2)-positive and triple-negative (TN) breast cancers and compare performance with dynamic contrast-enhanced (DCE)-MRI.

In this single-center study, patients who underwent NAC and surgery for HER2-positive or TN cancers between September 2020 and August 2021 were retrospectively selected to develop prediction models for pCR after NAC. From a prospective ASLAN (Avoid axillary Sentinel Lymph node biopsy After Neoadjuvant chemotherapy) trial, HER2-positive and TN cancer patients who underwent NAC and surgery between December 2021 and July 2022 were prospectively selected for model validation. Clinical-pathologic data and DM and MRI scans were obtained before and after NAC. Logistic regression analyses identified factors associated with pCR for model development and four models (clinical-pathologic, MRI, DM-AI-CAD, and combined) were evaluated.

A total of 259 women (mean age, 53 years ± 10.5 [SD]) constituted the development cohort and 119 (50.8 years ± 11.1) the validation cohort. Age, clinical N stage, estrogen receptor, progesterone receptor, and Ki-67 were incorporated into the clinical-pathologic model. In the validation cohort, the DM-AI-CAD model, applying AI-CAD score ≤ 16 on post-NAC DM as the radiologic CR criterion, showed a higher area under the receiver operating characteristic curve (AUC) compared to the clinical-pathologic model (0.72 vs. 0.62; p = 0.01) for pCR. However, the MRI model showed the highest AUC (0.83), then the combined model (0.78).

The model utilizing post-NAC DM with AI-CAD score ≤ 16 predicted pCR more accurately than the clinical-pathologic model in HER2-positive and TN cancers but was inferior to the MRI model.

Question The performance of digital mammography (DM) with AI-based computer-aided diagnosis (AI-CAD) for predicting pathologic complete response (pCR) after neoadjuvant chemotherapy (NAC) is unclear. Findings The DM-AI-CAD model incorporating AI-CAD score ≤ 16 on post-NAC DM predicted pCR more accurately than the clinical-pathologic model but not the MRI model. Clinical relevance The DM-AI-CAD model has potential to predict pCR after NAC in breast cancer patients for whom MRI is unavailable or contraindicated.