Prostate biopsy is commonly used to detect and stage prostate cancer. However, accurate targeting of biopsy requires the prior identification of appropriate targets for biopsy sampling. MRI-based decision support systems have significant potential to improve targeting by predicting which potential biopsy locations are most likely to have cancer, and which can safely be left unsampled. In this study, we developed an algorithm to predict whether a given biopsy would find clinically significant prostate cancer using MRI data alone. Using a dataset of 11,095 biopsies collected from 711 patients, we developed a support vector machine for predicting the presence of clinically significant prostate cancer with an AUC of 0.71 ± 0.02 on cross-validation with a precision of 0.86 ± 0.02 and a sensitivity of 0.72 ± 0.01.
@inproceedings{Shi2018,
author = {Shi, W. and Sarma, K. V. and Raman, A. G. and Priester, A. M. and Natarajan, S. and Speier, W. and Raman, S. S. and Marks, L. S. and Arnold, C. W.},
booktitle = {Medical Imaging Meets NeurIPS Workshop},
title = {{Prediction of Clinically Significant Prostate Cancer in MR/Ultrasound Guided Fusion Biopsy using Multiparametric MRI}},
year = {2018},
}