Pathology of transplanted heart rejection using artificial intelligence-based image analysis of endomyocardial biopsies

Introduction: Technologies based on digital image analysis are becoming an increasingly prominent feature of pathological diagnostics. The application of artificial intelligence to data analysis has the potential to offer a more objective and detailed morphological characterization than that achievable through visual inspection. This could lead to a reduction in the time necessary for a diagnosis to be reached. Objective: The aim of this study was to optimize the nuclear recognition and nuclear separation capabilities of the image analysis software BIAS (Single-Cell Technologies). Method: To this end, the recognition and morphological characteristics (distance, density) of five to five Gr0R, Gr1R, Gr2R stage endomyocardial biopsies of hematoxylin-eosin stained, digitized sections of lymphocytes, myocytes, and other tissue structures were investigated. Results: The data demonstrated a clear increase in lymphocyte density averages during the progression of histological signs of graft rejection (Gr0R: 127.02/mm(2 )< Gr1R: 324.03/mm2 < Gr2R: 686.49/mm2), with the results for Gr0R showing a significant difference compared to Gr1R. The mean distance between lymphocytes exhibited a corresponding variation (Gr0R: 32.44 am > Gr1R: 19.37 am > Gr2R: 11.63 am), with the latter two values being significantly below the Gr0R cases. The mean myocyte-lymphocyte distances of the first ten lymphocytes in order of distance from the myocytes were found to be similar (Gr0R: 55.32-193 am > Gr1R: 35.16-109.96 am > Gr2R: 32.46-92.95 am). This indicates that the mean distance of lymphocytes from myocytes in Gr0R cases was significantly greater than in the other groups. In 1 mm(2) of myocardium, the mass of intramyocardial connective tissue exhibited a notable decline following a substantial increase (Gr0R: 1013.72 am2, Gr1R: 1942.65 am2, Gr2R: 1686.79 am2). Conversely, the prevalence of intramyocardial oedema demonstrated an appreciable surge subsequent to a moderate decline (Gr0R: 202.42 am2, Gr1R: 181.56 am2, Gr2R: 273.91 am2) throughout the progression of the rejection process. Discussion: The results of our study indicate that our artificial intelligence-based method, when adequately trained, is suitable for objective pathological analysis of lymphocyte, myocyte and connective tissue volume, as well as the extent of oedema and morphological parameters (distance, density) that are important from the perspective of rejection in endomyocardial biopsies of transplanted hearts. Conclusion: Complex digital image analysis may prove to be a valuable tool for the efficient pathological evaluation of organ rejection in heart transplant recipients.