In recent years there has been an interest in multimodality data analysis for disease detection. Ideally, multimodal methods should leverage the strengths of each modality and compensate for weaknesses. With this abundance of data types come the issues of limited samples per modality, missing features, spatial registration of different modalities, and feature and classifier selection. In the first half of this talk, I will describe my experience with multiparametirc and multimodality data analysis for prostate cancer detection. This will cover conventional ultrasound imaging methodologies along with my innovations in ultrasound-based cancer detection using RF time series analysis. The second half of the talk will be a deep dive into the issue of handling datasets with a large number of samples with missing features. I describe the newly proposed concept of scandent trees. This is a novel random forest learning method for incomplete multimodal datasets with immediate applications in combining imaging and genomic data. I will show how this approach significantly improves the performance of a classifier designed for genomics plus imaging analysis by enabling the use of large amounts of archival imaging data.