Work

Radiomics analysis based on DBT and US for differentiating benign from malignant breast lesions: a multicenter study

Year: 2026

Type: article

Abstract: Accurate differentiation between benign and malignant breast lesions remains challenging in clinical practice due to the subjective interpretation of conventional imaging. Although digital breast tomosynthesis (DBT) and ultrasound (US) provide complementary diagnostic information, their integration remains limited, highlighting the need for quantitative and multimodal imaging approaches. This study aimed to develop and evaluate traditional radiomics, habitat, and deep learning feature-based models (DL models) based on DBT and US for breast lesion classification. A retrospective study was conducted at two centers: center 1 ( n = 183) and center 2 ( n = 65). Breast lesions were categorized using the Breast Imaging-Reporting and Data System (BI-RADS) to construct a clinical model. The region of interest (ROI) for DBT and US images was delineated by experienced radiologists. Radiomics and habitat features were extracted and selected to establish radiomics and habitat models. Deep features were extracted using a pretrained ResNet101, and a DL model was developed based on these extracted features. The Delong test was used to compare the area under the curve (AUC) among different models. In the training cohort, the clinical, habitat, and DL models achieved AUC values of 0.943 ( P < 0.001), 0.965 ( P = 0.003), and 0.926 ( P < 0.001), respectively, significantly outperforming the radiomics model (AUC = 0.826). No statistically significant differences were found among the clinical, habitat, and DL models ( P > 0.05). In the test cohort, the clinical model and the DL model significantly outperformed the radiomics model, with AUCs of 0.943 ( P < 0.001) and 0.874 ( P = 0.012), respectively. Multimodal imaging–based radiomics analysis shows promise as a noninvasive approach for differentiating benign from malignant breast lesions. The DL model demonstrated robust overall diagnostic performance in external validation, suggesting potential utility as an adjunctive tool to support lesion characterization and reduce unnecessary biopsies. These findings underscore the potential of combining DBT and US imaging analytics to advance precision breast imaging and support future prospective multicenter studies for clinical implementation. Not applicable.

Source: BMC Medical Imaging

DOI: 10.1186/s12880-026-02704-6

Authors: Dong-Man Ye, Si-Meng Song, Peng Lu, Chun-Xiao Cui, Kui-Ran Gao, Tao Yu

Institutions: Liaoning Cancer Hospital & Institute, Qingdao University, Affiliated Hospital of Qingdao University

Language: English

Topic: Radiomics and Machine Learning in Medical Imaging

Subfield: Radiology, Nuclear Medicine and Imaging

Field: Medicine

Domain: Health Sciences

SDG: Good health and well-being

Open Access status: gold

Estimated APC paid: $3,090