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

A computer system having one or more processors, and memory storing one or more programs for execution by the one or more processors, the one or more programs comprising instructions for performing a method of identifying a risk of kidney graft failure for a subject, the method comprising:

• obtaining, in electronic format, a first image of a kidney graft biopsy on a substrate, wherein the first image represents at least a first subset of morphological classes in a plurality of morphological classes comprising at least a first tissue compartment class and an inflammatory mediator class, the first image comprises at least 10,000 pixels, and the graft biopsy originates from the subject;
• inputting the first image into a trained model, wherein the trained model comprises at least 10,000 parameters;
• identifying, as output from the trained model, for each respective morphological class in the at least the first subset of morphological classes, a corresponding one or more objects in the first image that fall within the respective morphological class, wherein the first subset of morphological classes includes the first tissue compartment class or the inflammatory mediator class;
• generating, for each respective morphological class in the at least the first subset of morphological classes, a corresponding plurality of digital features comprising:
• (i) a first corresponding subset of individual feature scores, wherein each respective individual feature score is obtained using the corresponding one or more objects for the respective morphological class, and
• (ii) a second corresponding subset of composite feature scores, wherein each respective composite feature score is obtained by combining two or more individual feature scores in the first corresponding subset of individual feature scores;
• and
• comparing, for each respective morphological class in the at least the first subset of morphological classes, a respective digital feature in the corresponding plurality of digital features for the respective morphological class to a reference criterion, thereby determining the risk of kidney graft failure for the subject.

A non-transitory computer readable storage medium storing one or more programs configured for execution by a computer, the one or more programs comprising instructions for carrying out a method for identifying a risk of kidney graft failure for a subject, the method comprising:

• obtaining, in electronic format, a first image of a kidney graft biopsy on a substrate, wherein the first image represents at least a first subset of morphological classes in a plurality of morphological classes comprising at least a first tissue compartment class and an inflammatory mediator class, the first image comprises at least 10,000 pixels, and the graft biopsy originates from the subject;
• inputting the first image into a trained model, wherein the trained model comprises at least 10,000 parameters;
• identifying, as output from the trained model, for each respective morphological class in the at least the first subset of morphological classes, a corresponding one or more objects in the first image that fall within the respective morphological class, wherein the first subset of morphological classes includes the first tissue compartment class or the inflammatory mediator class;
• generating, for each respective morphological class in the at least the first subset of morphological classes, a corresponding plurality of digital features comprising:
• (i) a first corresponding subset of individual feature scores, wherein each respective individual feature score is obtained using the corresponding one or more objects for the respective morphological class, and
• (ii) a second corresponding subset of composite feature scores, wherein each respective composite feature score is obtained by combining two or more individual feature scores in the first corresponding subset of individual feature scores;
• and
• comparing, for each respective morphological class in the at least the first subset of morphological classes, a respective digital feature in the corresponding plurality of digital features for the respective morphological class to a reference criterion, thereby determining the risk of kidney graft failure for the subject.