Structural image processing
To exploit important and potentially prognostic tissue image characteristics, we include DWI and FLAIR image processing into our framework. To this end, we develop an unsupervised deep learning method to create a quantitative imaging “biomarker”, implicative of DWI-FLAIR mismatch. To show validity, we conduct visual assessment and to analyze prognostic value of the biomarker we utilize Neural Networks to predict clinical outcomes. We hypothesize an improved prognosis of the estimated mismatch biomarker compared to the hard thresholding technique of manual inspection.