AI accurately predicts malignancy on breast ultrasound, cuts down excessive follow-ups and biopsies

The study revealed that an AI method showed comparable performance to that of radiologists and can help avoid unnecessary biopsies and follow-up exams, which, in turn, can contribute to sustainability in healthcare practices.

“By considering AI-assigned BI-RADS 2 as safe, we could potentially avoid 11% of benign lesion biopsies and 46.2% of follow-ups,” Nilgun Guldogan, Breast Clinic, Acibadem Altunizade Hospital, Istanbul, Turkey, and colleagues reported.

Previous studies have shown increased use of artificial intelligence systems in breast ultrasonography. These studies have shown how AI aids in image interpretation, reduces false-positive cases, and potentially helps decrease the workload of radiologists. Therefore, Dr Guldogan and colleagues evaluated the performance of an AI system for the BI-RADS category assessment in breast masses detected on breast ultrasound.

For this purpose, the researchers analyzed 715 masses detected in 530 patients. Nine breast radiologists and three breast imaging centres of the same institution participated in the study. One radiologist performed an ultrasound and obtained two orthogonal views of each detected lesion. A second radiologist blinded to the patient’s clinical data retrospectively reviewed these images. The images were evaluated by a commercial AI system.

The level of agreement between the AI system and the two radiologists and their diagnostic performance were calculated according to dichotomic BI-RADS category assessment.

The study revealed the following findings:

· The study included 715 breast masses. Of these, 18.75% were malignant, and 81.25% were benign.

· The agreement between AI and the first and second radiologists was moderate statistically in discriminating benign and probably benign from suspicious lesions.

· The sensitivity and specificity of radiologist 1, radiologist 2, and AI were calculated as 98.51% and 80.72%, 97.76% and 75.56%, and 98.51% and 65.40%, respectively.

· For radiologist 1, the positive predictive value (PPV) was 54.10%, the negative predictive value (NPV) was 99.58%, and the accuracy was 84.06%.

· Radiologist 2 achieved a PPV of 47.99%, NPV of 99.32%, and accuracy of 79.72%.

· The AI system exhibited a PPV of 39.64%, NPV of 99.48%, and accuracy of 71.61%.

· None of the lesions categorized as BI-RADS 2 by AI were malignant, while 2 of the lesions classified as BI-RADS 3 by AI were subsequently confirmed as malignant.

· By considering AI-assigned BI-RADS 2 as safe, 11% of benign lesion biopsies and 46.2% of follow-ups could be potentially avoided.

"Artificial intelligence proves effective in predicting malignancy," the researchers wrote, they added, "Integrating it into the clinical workflow has the potential to reduce short-term follow-ups and unnecessary biopsies, which, in turn, can contribute to sustainability in healthcare practices."

Reference:

Guldogan, N., Taskin, F., Icten, G. E., Yilmaz, E., Turk, E. B., Erdemli, S., Parlakkilic, U. T., Turkoglu, O., & Aribal, E. (2023). Artificial Intelligence in BI-RADS Categorization of Breast Lesions on Ultrasound: Can We Omit Excessive Follow-ups and Biopsies? Academic Radiology. https://doi.org/10.1016/j.acra.2023.11.031