What’s beyond breast asymmetry? Comparative study between artificial intelligence and contrast-enhanced spectral mammography in the assessment of various types of breast asymmetries

Abstract

BackgroundBreast asymmetries are prevalent findings in mammograms and are commonly caused by variations in normal breast tissue. However, they may imply significant underlying causes in some cases. Such cases necessitate further assessment by adding further mammography views, targeted ultrasound, and investigations to exclude underlying pathology.ObjectivesTo investigate the role of artificial intelligence (AI) compared to contrast-enhanced spectral mammography (CESM) in the assessment of breast asymmetries and their performance as diagnostic modality among different types of breast asymmetries as well as the additive value of AI software to mammography in these cases.MethodsSixty-four female patients were diagnosed with breast asymmetries by standard mammography (MMG) on both craniocaudal (CC) and mediolateral oblique (MLO). Digital breast tomosynthesis (DBT) may have been added. After evaluating the breast asymmetry by MMG and complementary breast ultrasound (US), both CESM and AI were performed for all cases and analyzed, then the interpreted results were compared accordingly either by histopathology from suspected lesions scored as BI-RADS 4 or 5 or through further close follow-up by single-view mammography in benign cases scored as BI-RADS 2 or 3.ResultsThe sensitivity and specificity of CESM in the assessment of breast asymmetry in correlation with pathological data/follow-up results were 100% and 60% (p < 0.001). The corresponding values for AI were 70.83% and 75%; however, the estimated overall accuracy for both CESM and AI was close to each other measuring 75% and 73.44%, respectively (p < 0.001). The diagnostic accuracy of CESM to detect malignant causes of breast asymmetry was 100%; however, the detection of benign causes of breast asymmetry was 40%. The corresponding values for AI were 70.83% and 25%, respectively, with significant p-value (p < 0.001).ConclusionsThe CESM was more sensitive; however, the AI was more specific in the assessment of different breast asymmetries. Although the diagnostic accuracy of both is close to each other. Therefore, AI-aided reading can replace CESM in most cases, especially for those contraindicated to do CESM. AI also can reduce the radiation exposure hazards of a second dose of radiation for CESM and its financial cost as well. AI-aided reading in breast screening programs can reduce the recall of patients, unnecessary biopsies, and short-interval follow-up exams.