To demonstrate the added predictive value of radiomic features to prostate radiology scoring scheme (PIRADS), a systematic approach is required to determine whether there is indeed latent predictive information of prostate cancer in diffusion-weighted magnetic resonance images (DW-MRI) that cannot be captured by radiologists’ visual interpretations alone. In this work, we propose a PI-RADS guided discovery radiomics solution where a predictive model for prostate cancer is built by discovering radiomic features that capture information on the phenotype of lesions, which is not visible to radiologists when using PI-RADS scoring system. We investigated patients with PI-RADS scores indicating presence or absence of significant prostate cancer separately and ran experiments on patients with DW-MRI followed by targeted biopsy, using first and second order quantitative imaging features. Our experiments on DW-MRI and pathology data of 50 patients show that the proposed approach improves the overall accuracy of prostate cancer diagnosis significantly compared to PI-RADS scores alone.

Date Published: 13 March 2019
PDF: 6 pages
Proc. SPIE 10950, Medical Imaging 2019: Computer-Aided Diagnosis, 1095042 (13 March 2019); doi: 10.1117/12.2512550

Published in SPIE Proceedings Vol. 10950:
Medical Imaging 2019: Computer-Aided Diagnosis
Kensaku Mori; Horst K. Hahn, Editor(s)