Advanced Morphologic Analysis for Diagnosing Allograft Rejection

Abstract

Allograft rejection remains a significant concern following all solid organ transplants. While qualitative morphologic analysis with histologic grading of biopsy samples is the main tool employed, for diagnosing allograft rejection, this standard has significant limitations in precision and accuracy that affect patient care. The use of endomyocardial biopsy (EMB) to diagnose cardiac allograft rejection (CAR) illustrates the significant shortcomings of current approaches for diagnosing allograft rejection. Despite disappointing interobserver variability, concerns about discordance with clinical trajectories, attempts at revising the histologic criteria and efforts to establish new diagnostic tools with imaging and gene expression profiling, no method has yet supplanted EMB as the diagnostic gold standard. In this context, automated approaches to complex data analysis problems – often referred to as ‘machine learning’ – represent promising strategies to improve overall diagnostic accuracy. By focusing on cardiac allograft rejection, where tissue sampling is relatively frequent, this review highlights the limitations of the current approach to diagnosing allograft rejection, introduces the basic methodology behind machine learning and automated image feature detection, and highlights the initial successes of these approaches within cardiovascular medicine.

Cell-free DNA donor fraction analysis in pediatric and adult heart transplant patients by multiplexed allele-specific quantitative PCR: Validation of a rapid and highly sensitive clinical test for stratification of rejection probability

“…EMB is also hampered by alarmingly low inter-observer (pathologist) concordance in assignment of rejection grade despite modifications in grading criteria [20][21][22][23]. In addition, the incidence of "biopsy negative" rejection evidenced by hemodynamic compromise without demonstrable myocardial inflammation remains at approximately 20% [23][24][25][26]. It is thus a problematic gold standard, and some centers seek to reduce incidence of its use, particularly in infants and children, but also in adults after the first year post-transplant [27].…”

Deep Neural Network-Aided Histopathological Analysis of Myocardial Injury

“…While the high attention regions are correlated with pathological patterns, these patterns are not explicitly defined and cannot be quantified by DNNs. We will further introduce additional labels to characterize specific pathological patterns, for example, infiltrated inflammation and myocardial necrosis ( 12 ). Second, our data were provided by a single institute.…”

“…Machine learning methods have been an integral part of biomedical research (5,6) and clinical work (7,8), having the great potential to overcome the intra-and inter-observer variability (9,10) and to improve diagnostic accuracy and efficiency (11). These computational models are based on algorithms that can extract features from clinical data (12). Compared to traditional machine learning methods that rely on expert knowledge to transform raw image data into features (e.g., texture, statistics, and wavelet transform coefficients) (13,14), deep neural networks (DNN) can achieve better accuracy without defining features explicitly.…”

Molecular Diagnosis of Rejection in Heart Transplantation

“…These automated approaches seek to overcome subjective differences in histopathologic interpretation, and potentially reduce interobserver variability in diagnosis. 39 Preliminary results for the ability of machine learning to successfully discriminate differentiate between grades of ACR in heart transplant recipients have been promising. One such tool, the CACHE-Grader, demonstrated that automated histologic diagnosis is feasible and noninferior to evaluation by expert pathologists.…”