Identifying breast cancer subtypes can go a long way in clinically managing the disease. Certain molecular subtypes respond well to therapy and the information can prove to be invaluable on the path to survival.
In a study published online March 8 in Academic Radiology, researchers used a machine learning radiomic technique to extract quantitative imaging features from digital mammograms. The platform demonstrated the imaging features were associated with breast cancer subtypes.
U.S. and Chinese scientists retrospectively analyzed mammograms of 331 Chinese women diagnosed with invasive breast cancer in 2015. Each patient had both the craniocaudal (CC) view and the mediolateral oblique (MLO) view images.
The group consisted of 29 triple-negative, 45 human epidermal growth factor receptor 2 (HER2)-enriched, 36 luminal A and 221 luminal B lesions.
Radiomic features were taken from the lesion areas of each image—resulting in 39 total features. Authors performed three binary classifications of the lesion subtypes. Triple-negative versus non-triple-negative, HER2-enriched versus non-HER2-enriched and luminal (A and B) versus nonluminal.
A machine algorithm classified the findings into subtypes. The combined CC/MLO technique achieved an accuracy of 0.865 for triple-negative vs. non-triple negative, 0.784 for HER2-enriched vs non HER2 enriched and 0.752 for luminal vs nonluminal subtypes.
The 12 most predictive features were determined by the least absolute shrink age and selection operator method. Four features (roundness, concavity, gray mean and correlation) were determined to be the most significant identifiers of subtypes.
“Our results on the three binary classifications of subtypes (i.e., triple-negative vs other types, HER2-enhanced vs other types, and luminal vs other types) showed that a set of such quantitative radiomic features is predictive of the molecular subtypes of breast cancer,” wrote corresponding authors Peifang Liu, PhD, and Shandong Wu, PhD, both with the department of breast imaging at Tianjin Medical University Cancer Institute and Hospital’s National Clinical Research Center for Cancer in China, and colleagues.
Authors stressed their pilot study needed to be tested in larger studies in the future.
“If the automated radiomic features like we identified in this study are validated to be predictive of the molecular subtypes, it can provide further information from the images to aid radiologists in mammographic reading and to better inform clinical diagnosis and decision-making,” wrote Liu and Wu et al. “This would have important additional value too for patients who do not have a breast MRI scan available.”