The Clinical Application of Artificial Intelligence in Breast Imaging: Current Insights, Challenges, and Future Directions

Abstract

Artificial intelligence (AI) is used in various areas of radiology, particularly in breast imaging, starting with mammography and extending to ultrasonography (US) and magnetic resonance imaging (MRI). This overview aims to examine the introduction, applications, and challenges of AI in breast imaging. This narrative outlines the applications of AI in various modalities-including mammography, US, and MRI-and discusses its indications, ongoing challenges, and future perspectives. AI has been used for identification, classification, detection, diagnosis, breast density assessment, treatment response, and prediction of prognosis. AI can help radiologists avoid missed diagnoses due to heavy workloads and enhance workflow efficiency. The integration of AI software into daily practice, along with further validation and refinement, is necessary to support radiologists' workflows.

Keywords: Artificial Intelligence; Breast; Magnetic Resonance Imaging; Mammography; Ultrasonography.

Figure 1. A 32-year-old woman who presented for diagnostic evaluation due to a palpable mass in her left breast. (A, B) Left mammography images in the craniocaudal (A) and mediolateral oblique (B) views show no visible lesion at the palpable site, marked with a BB marker. (C) The artificial intelligence algorithm highlights the left mediocentral area corresponding to the palpable site with two circular marks, displaying an abnormality score of 99%.

Figure 2. A 41-year-old woman with pathologically confirmed fibroadenoma. (A) She had a 2.2 cm mass on B-mode sonography. (B) The system fully automatically measured the size as 2.1 × 1.2 cm and analyzed the lesion with lexicon descriptors of oval shape, parallel orientation, circumscribed margin, and hypoechoic echo pattern. CadAI-B highlighted a mass and generated a differential diagnosis report of “probably benign” (Breast Imaging-Reporting And Data Systems 3), with a CadAI-score of 3 out of 100 (by courtesy of Professor Wonhwa Kim and Professor Jaeil Kim).

Figure 3. A 47-year-old woman with pathologically confirmed invasive ductal carcinoma. (A) She had a 1.4 cm mass on B-mode sonography. (B) The system fully automatically measured the size as 1.4 × 0.9 cm and analyzed the lesion with lexicon descriptors of irregular shape, parallel orientation, non-circumscribed margin, and hypoechoic echo pattern. CadAI-B highlighted a mass and generated a differential diagnosis report of “moderate suspicion for malignancy” (Breast Imaging-Reporting And Data Systems 4B), with a CadAI-score of 44 out of 100 (by courtesy of Professor Wonhwa Kim and Professor Jaeil Kim).