Solutions for Missing Parameters in Computer-Aided Diagnosis with Multiparametric Imaging Data

Multiparametric MRI (mpMRI) is becoming widely used as a means of determining the need for prostate biopsy and also for targeting prostate biopsies. One problem with the mpMRI approach is that not all MRI modalities might be available for each patient. For example, the use of gadolinium-based contrast agents in dynamic contrast enhanced MRI (DCE-MRI) results in allergic reactions in some patients with reported reaction rates as high as 19.8% which results in missing DCE-MRI parametric maps. The process of modifying a classifier to work on incomplete dataset is challenging and time consuming. This modification may require a time consuming retraining or having multiple classifiers for each missing data type. Therefore, the objective of the work presented here is to develop an image-based classification technique for the detection of prostate cancer with the capability of handling missing DCE parameters. We propose four different methods and show their effectiveness in maintaining high Area Under Curve (AUC) while handling missing parameters without the requirement of any modifications to the classifier models.