Deep learning approach identified pathological abnormalities predictive of graft loss in kidney transplant

Abstract

Systems and methods for identifying risk of kidney graft failure in a subject are provided. An image of a graft biopsy is obtained, the image representing at least a subset of a plurality of morphological classes including a tissue compartment and inflammation class. The image is inputted into a trained model that identifies, as output, for each morphological class in at least the subset, corresponding objects in the image that fall within the morphological class. For each morphological class, corresponding digital features are generated comprising a first subset of individual feature scores obtained using the objects for the morphological class and a second subset of composite feature scores obtained by combining individual feature scores. Digital features for each morphological class are compared to a reference criterion, thus determining risk of kidney graft failure.