Satish Viswanath1, B. Nicolas Bloch2,
Jonathan Chappelow1, Pratik Patel1, Neil Rofsky3,
Robert Lenkinski4, Elisabeth Genega4, Anant Madabhushi1
1Biomedical Engineering, Rutgers
University, Piscataway, NJ, United States; 2Boston Medical Center;
3UT Southwestern Medical Center; 4Beth Israel Deaconess
Medical Center
We present a novel technique, Enhanced Multi-Protocol Analysis via Intelligent Supervised Embedding (EMPrAvISE), for building a computerized meta-classifier to predict the spatial extent of prostate cancer (CaP) in vivo via multi-protocol (T2-weighted, Dynamic Contrast Enhanced, Diffusion-weighted) MRI data. We employ automated registration, quantitative image descriptors, and a novel ensemble representation technique in our methodology. Evaluation of our automated predictions for spatial extent of CaP at a pixel level (against registered extents of CaP on MRI) results in EMPrAvISE showing a statistically significant improvement (AUC=0.73) over individual protocols (T2w, DCE,DWI), as well as simple multi-protocol feature concatenation.