AI Model Accurately Predicts Malignancy on Breast Ultrasound
Posted on 22 Dec 2023
Artificial intelligence (AI) systems are increasingly being integrated into breast ultrasonography to potentially reduce radiologists' workload and enhance diagnostic precision. Now, a new study that evaluated an AI system's performance in BI-RADS category assessment for breast masses detected on ultrasound has found that the technology can effectively predict malignancy.
The study, conducted at Acibadem Altunizade Hospital (Istanbul, Turkey), involved the analysis of 715 masses across 530 patients. It engaged three breast imaging centers from the same institution and nine breast radiologists. Ultrasound examinations were carried out by one radiologist capturing two orthogonal views of each lesion. These images were then retrospectively examined by a second radiologist who was not privy to the patient’s clinical information. A commercially available AI system also evaluated the images. The researchers measured the level of concordance between the AI system and the radiologists, along with their diagnostic effectiveness, according to the dichotomous BI-RADS category assessment.
The study noted a moderate level of concordance between the AI model and the radiologists in differentiating benign and probably benign lesions from those deemed suspicious. The AI model ascertained that none of the lesions categorized as BI-RADS 2 were malignant, although two classified as BI-RADS 3 were confirmed malignant. The researchers highlighted that considering BI-RADS 2 lesions identified by AI as non-threatening could allow radiologists to avoid numerous unnecessary benign lesion biopsies and a significant number of follow-ups. Additionally, the AI algorithm potentially downgraded a considerable percentage of BI-RADS 3, 4, and 5 lesions to BI-RADS 2 or 3 and upgraded numerous benign or possibly benign lesions as suspicious, albeit with a low malignancy risk. The researchers concluded that AI holds promise in accurately predicting malignancy, and its integration into clinical workflows could reduce unnecessary biopsies and follow-ups, thereby enhancing sustainability in healthcare practices.
Related Links:
Acibadem Altunizade Hospital
