Development of A Radiomics Signature Using CT Texture Analysis of Breast Lesions Secondary to Infiltrating Ductal Carcinoma to Predict Histologic Features: A Cross-Sectional Study

Rahim, Anila; Awais, Muhammad; Hafeez, Mahnoor; Rehman, Abdul; Sattar, Amjad

doi:10.31557/APJCP.2026.27.10.3829

Development of A Radiomics Signature Using CT Texture Analysis of Breast Lesions Secondary to Infiltrating Ductal Carcinoma to Predict Histologic Features: A Cross-Sectional Study

Document Type : Research Articles

Authors

¹ Department of Radiology, Dow University of Health Sciences, Karachi, Sindh, Pakistan.

² Department of Radiology, Aga Khan University Hospital, Karachi, Sindh, Pakistan.

³ Department of Medicine, TidalHealth Peninsula Regional, Salisbury, MD, United States of America.

10.31557/APJCP.2026.27.10.3829

Abstract

Objective: Radiomics has emerged as a promising approach for the non-invasive characterization of tumor biology using quantitative imaging features. Most radiomics studies in breast cancer have focused on magnetic resonance imaging (MRI), while the potential utility of computed tomography (CT)-based texture analysis remains underexplored. This study aimed to develop radiomic signatures derived from CT texture analysis of breast lesions to predict histologic features. Methods: Patients with biopsy-proven infiltrating ductal carcinoma (IDC) who had undergone contrast-enhanced chest CT less than 1 month prior to biopsy at Dow University of Health Sciences during the study period (January 2023 to December 2023) were included. A region of interest (ROI) was manually drawn around the primary breast lesion, and radiomic features were extracted using LIFEx® software. An elastic-net regression model was used to select radiomic features for the development of the final radiomics signature. Results: Ninety-two women with biopsy-proven IDC were included (median age, 41.5 years; IQR, 32.7–51.0). Radiomic signatures demonstrated fair predictive performance for several histologic features. The model predicting high-grade tumors achieved an area under the receiver operating characteristic curve (AUROC) of 0.763, with an accuracy of 71.7%. For a high acinar score, the AUROC was 0.706, with an accuracy of 66.3%. The radiomic signature for high nuclear grade showed the best performance, with an AUROC of 0.827 and an accuracy of 77.2%. Prediction of PR expression yielded an AUROC of 0.731 and an accuracy of 70.6%. No reliable radiomic signatures could be developed for mitotic rate, estrogen receptor expression, HER2/neu status, or Ki-67 proliferation index. Conclusion: In this proof-of-concept study, radiomic signatures based on CT texture analysis of IDC allowed prediction of histologic grade (nuclear grade and acinar score) and PR status with fair diagnostic accuracy. Further validation of these findings in larger cohorts of women with varying breast densities and scanned using different types of CT scanners is warranted.

Keywords

Main Subjects