UCLA

Prostate cancer is one of the leading causes of death due to cancer for men in the United States. Current diagnostic procedures have been shown to lead to overdiagnosis and overtreatment of the disease. Thus, a need exists for diagnostic protocols with higher positive predictive value (PPV). Magnetic resonance imaging (MRI) guided biopsy has emerged as a new diagnostic technology that has increased the ability to detect cancerous tissue in the prostate. In this study, we utilize a dataset of 555 patients who have undergone MRI-guided biopsy to answer two questions: 1) how accurate are radiologist drawn regions of interest (ROIs) on prostate MRI, and 2) can we map the location of a biopsy to an MRI and use MRI voxel intensities at that location to predict whether or not the biopsy core contains clinically significant cancer (csCaP). In answering the first question, we found that 50.35% of csCaP-containing cores are found inside ROIs while 49.65% of csCaP-containing cores are found outside of ROIs. This indicates room for improvement in ROI delineation. We then trained support vector machine (SVM) and logistic regression classifiers using features from the MRI voxels corresponding to each biopsy core’s location to predict whether cores would be cancer-positive or cancer-negative. The SVM achieved the best performance, with a negative predictive value of 0.93, a PPV of 0.23, and a test area under the curve (AUC) of 0.72.