AI improves with input from radiologists
Artificial intelligence (AI) models utilizing radiologist-provided BI-RADS classification outperformed methods that did not use them, according to an Oct. 15 study in the Journal of the American College of Radiology.
Despite radiology’s widespread interest in AI, research investigating radiologist-augmented approaches has not received the same attention as algorithms which operate independently of humans, argued Adarsh Ghosh, MD with the Department of Radiodiagnosis and Imaging at AIIMS in New Delhi, India.
Ghosh obtained BI-RADS data sets from the University of California, Irvine Machine Learning Repository (dataset 1) and the Digital Database for Screening mammography repository (dataset 2).
The team trained two sets of models: M1, using lesion shape, margin, density, and patient age for data set 1 and image texture parameters for data set 2; and M2, using the same parameters along with BI-RADS classification completed by radiologists.
Compared to the validation data set, Ghosh reported the BI-RADS classification model performed “significantly” better than the M1 model which did not use such data.
“Though the parameters used in these models are very simplistic and the data size is small, the results have successfully demonstrated that a radiologist-augmented workflow is feasible in AI, allowing better management of patients and disease classification,” Ghosh concluded. “Further evaluation of the metrics involved in predictor handling by AI algorithms will provide newer insights into imaging.”
Matt joined Chicago’s TriMed team in 2018 covering all areas of health imaging after two years reporting on the hospital field. He holds a bachelor’s in English from UIC, and enjoys a good cup of coffee and an interesting documentary.