Computerized Image Analysis to Differentiate Benign and Malignant Breast Tumors on Magnetic Resonance Diffusion Weighted Image: A Preliminary Study

Purpose This work aims to determine the feasibility of using a computer-aided diagnosis system to differentiate benign and malignant breast tumors on magnetic resonance diffusion-weighted image (DWI). Materials and Methods Institutional review board approval was obtained. This retrospective study included 76 patients who underwent breast magnetic resonance imaging before neoadjuvant chemotherapy from March 10, 2017, to October 12, 2017, with a total of 80 breast tumors including 40 cases of breast cancers and 40 cases of benign breast tumors. The textural features of DWI images were analyzed. The area under the receiver operating characteristic curve was calculated to evaluate the diagnostic efficiency of texture parameters. Multiple linear regression analysis was used to determine the efficiency of texture parameters for distinguishing the 2 types of breast tumors. Results Computer vision algorithms were applied to extract 67 imaging features from lesions indicated by a breast radiologist on DWI images. A total of 19 texture feature parameters, such as variance, standard deviation, intensity, and entropy, out of 67 texture parameters were statistically significant in the 2 sets of data ( P < 0.05). By comparing the receiver operating characteristic curves, we found that the mean and relative deviations exhibited high diagnostic values in differentiating between benign and malignant tumors. The accuracy of Fisher discriminant analysis for the 2 types of breast tumors was 92.5%. Conclusions Breast lesions exhibit certain characteristic features in DWI images that can be captured and quantified with computer-aided diagnosis, which enables good discrimination of benign and malignant breast tumors.