Cardiac transplant patients must be on lifelong surveillance for the occurrence of acute cellular rejection (ACR). Despite potent immunosuppressants, ACR occurs when recipient T cells recognize donor antigens to cause cardiac myocyte destruction. Identifying myocyte damage is critical for determining the grade of rejection, and thus greatly impacts treatment decisions. The aim of this study is to determine if a machine learning algorithm can reach acceptable performance of identifying myocyte damage representative of ACR from benign conditions. AI technology has recently been under robust investigation in cancer pathology but is in its infancy in transplant pathology. This is the first AI algorithm to our knowledge, to identify myocyte damage in cardiac transplant patients. All regions of myocyte damage were annotated by a board-certified cardiac pathologist at a high-volume cardiac transplant center. Cases diagnosed as antibody-mediated rejection (AMR) were excluded. Annotations were extracted using the Openslide software and fed into a convolutional neural network which used the locked VGG16 base for transfer learning. Keras was used for AI modeling and training. Four annotation classes were included in the algorithm (normal, healing injury, Grade 1R1A and Grade 1R2). A total of 19,617 annotations (10,855 regions of ACR, 5002 healing injury, 3760 normal) were completed from 200 hematoxylin and eosin slides scanned using a Leica Aperio AT2 digital whole slide scanner at 40X magnification. The AI algorithm successfully distinguished myocyte damage (Grade 1R2) from non-myocyte damage (Grade 1R1A) with 94% validation accuracy. ACR was distinguished from healing injury with 98-99% validation accuracy, and from normal cardiac myocytes with 99% validation accuracy. This study provides evidence for the first time, that a machine learning algorithm can be taught and validated to distinguish myocyte damage ACR from non-myocyte damage ACR. To our knowledge, this is one of the first AI tools developed in cardiac transplant pathology that may be used as an adjunct tool to decrease inter-observer diagnostic variability and reduce diagnostic error amongst pathologists.