An automated computational image analysis pipeline for histological grading of cardiac allograft rejection

Abstract

Aim: Allograft rejection is a serious concern in heart transplant medicine. Though endomyocardial biopsy with histological grading is the diagnostic standard for rejection, poor inter-pathologist agreement creates significant clinical uncertainty. The aim of this investigation is to demonstrate that cellular rejection grades generated via computational histological analysis are on-par with those provided by expert pathologists.

Methods and results: The study cohort consisted of 2472 endomyocardial biopsy slides originating from three major US transplant centres. The 'Computer-Assisted Cardiac Histologic Evaluation (CACHE)-Grader' pipeline was trained using an interpretable, biologically inspired, 'hand-crafted' feature extraction approach. From a menu of 154 quantitative histological features relating the density and orientation of lymphocytes, myocytes, and stroma, a model was developed to reproduce the 4-grade clinical standard for cellular rejection diagnosis. CACHE-grader interpretations were compared with independent pathologists and the 'grade of record', testing for non-inferiority (δ = 6%). Study pathologists achieved a 60.7% agreement [95% confidence interval (CI): 55.2-66.0%] with the grade of record, and pair-wise agreement among all human graders was 61.5% (95% CI: 57.0-65.8%). The CACHE-Grader met the threshold for non-inferiority, achieving a 65.9% agreement (95% CI: 63.4-68.3%) with the grade of record and a 62.6% agreement (95% CI: 60.3-64.8%) with all human graders. The CACHE-Grader demonstrated nearly identical performance in internal and external validation sets (66.1% vs. 65.8%), resilience to inter-centre variations in tissue processing/digitization, and superior sensitivity for high-grade rejection (74.4% vs. 39.5%, P < 0.001).

Conclusion: These results show that the CACHE-grader pipeline, derived using intuitive morphological features, can provide expert-quality rejection grading, performing within the range of inter-grader variability seen among human pathologists.

Keywords: Allograft rejection; Digital pathology; Heart transplant; Machine learning; Image analysis.

Overview of the ‘Computer-Assisted Cardiac Histologic Evaluation-Grader’ multicentre validation experiment. Nearly 2500 clinical transplant endomyocardial biopsy slides from three transplant centres were used to develop and validate the Computer-Assisted Cardiac Histologic Evaluation-Grader, an automated histological analysis pipeline for assigning standard-of-care cellular rejection grades. The Computer-Assisted Cardiac Histologic Evaluation-Grader performance was compared to both the grade of record and to independent pathologists performing re-grading, demonstrating non-inferiority to expert pathologists, generalizability to external datasets, and excellent sensitivity and negative predictive value.

Figure 1

Flowchart outlining the distribution of study histology slides by cohort of origin and experiment.

Figure 2

Summary of Computer-Assisted Cardiac Histologic Evaluation-Grader pipeline development, including image analysis, feature selection, model training, and grading performance validation. The Computer-Assisted Cardiac Histologic Evaluation-Grader pipeline involves computerized identification and multi-parameter quantitation of immune cell infiltrates, measuring size, counts, densities, and spatial relationships (panels 1–4, see Figure 3 for more detailed image feature extraction workflow examples). Following feature extraction (panel 5), features were assessed and ranked based on ability to discriminate between International Society for Heart and Lung Transplantation histological grades (panel 6). In the descending order of discriminatory ability, features were added to the Computer-Assisted Cardiac Histologic Evaluation-Grader predictive models until optimal performance was achieved on the training sets. These features were subsequently ‘locked down’ into the final Computer-Assisted Cardiac Histologic Evaluation-Grader models which were deployed in the study validation set (panel 7).

Figure 3

Computer-Assisted Cardiac Histologic Evaluation-Grader feature extraction approach. (A) Workflow for compartment segmentation. Left panel: a digitized clinical histology slide from EMB tissue stained in haematoxylin and eosin. Middle-left: K-means segmentation into myocytes (dark grey), interstitium/stroma (light grey), and non-myocyte nuclei (white). Middle-right: Dilated myocyte mask created from myocyte segmentation, identifying the myocardial compartment. Right: Overlay of myocardial mask onto original tile, demonstrating the ability to independently analyse lymphocytes within the myocardial vs. endocardial compartments (which in this example contains a Quilty lesion rather than an infiltrating lymphocyte focus). (B) Workflow for lymphocyte foci identification. Left panel: green = lymphocytes identified as ‘clustering’ together via area-thresholding of individual lymphocyte nuclei (overlay of individual lymphocytes in green which comprise a cluster). Middle panel: blue outline = Distinct lymphocyte clusters identified for feature extraction. Right panel: Applying proximity graph thresholding to lymphocyte clusters allows merging of nearby clusters into a common ‘lymphocyte focus’ for reproducing the foci counting as outlined in the International Society for Heart and Lung Transplantation histological rejection grading scheme.

Figure 4

By-grade examples of image analysis results: The first row shows biopsy specimen of different rejection grades. The second row demonstrates the proximity graphs (i.e. foci) in green built across the tissue over lymphocytes. The third and fourth rows demonstrate that the clusters of proximally situated lymphocytes are different in shape and size between high and low rejection grades and also between different grades. One may appreciate that in 2R and 3R cases (high-grade rejection cases), the lymphocyte clusters covered most of the tissue while in 0R and 1R (low-grade rejection cases), lymphocyte clusters are dispersed, small, and they cover only a small proportion of the tissue specimen.

Figure 5

Computer-Assisted Cardiac Histologic Evaluation-Grader validation results: confusion matrices and Receiver-operating characteristic curves demonstrating M₁ and M₂ performance on independent validation sets (S₁, S₂, and S₃).