The feasibility of model-free machine learning with variable features of MRI images on the prostate cancer application