Machine learning-supported interpretation of kidney graft elementary lesions in combination with clinical data

Abstract

Interpretation of kidney graft biopsies using the Banff classification is still heterogeneous. In this study, extreme gradient boosting classifiers learned from two large training datasets (n = 631 and 304 cases) where the “reference diagnoses” were not strictly defined following the Banff rules but from central reading by expert pathologists and further interpreted consensually by experienced transplant nephrologists, in light of the clinical context. In three external validation datasets (n = 3744, 589, and 360), the classifiers yielded a mean ROC curve AUC (95%CI) of: 0.97 (0.92–1.00), 0.97 (0.96–0.97), and 0.95 (0.93–0.97) for antibody‐mediated rejection (ABMR); 0.94 (0.91–0.96), 0.94 (0.92–0.95), and 0.91 (0.88–0.95) for T cell–mediated rejection; >0.96 (0.90–1.00) with all three for interstitial fibrosis–tubular atrophy. We also developed a classifier to discriminate active and chronic active ABMR with 95% accuracy. In conclusion, we built highly sensitive and specific artificial intelligence classifiers able to interpret kidney graft scoring together with a few clinical data and automatically diagnose rejection, with excellent concordance with the Banff rules and reference diagnoses made by a group of experts. Some discrepancies may point toward possible improvements that could be made to the Banff classification.

The Molecular Phenotype of Kidney Transplants: Insights From the MMDx Project

“…Histologic biopsy assessment usually involves assessment of lesions, DSA, and C4d in step 1, and interpretation by guidelines step 2, 76 and has limitations in reproducibility. 60 Variability can be reduced by having an "ensemble" of observers independently read the biopsy 77 and using some form of averaging, but this is not the usual practice and is not specified in the Banff guidelines.…”

Histologic and molecular features of antibody-mediated rejection

“…Another refinement is proposed by Kikic et al [12 ▪ ] who showed that a continuous scoring system of individual glomerular capillaries is superior to ordinal Banff cg scores to assess progression of transplant glomerulopathy, as validated by a better prediction of graft survival. Similar to efforts in pathologic classification, recently published data on machine learning-supported interpretation of histology lesions, which is combined with clinical data, is reported by Labriffe et al [13 ▪ ]. They used a diagnostic classifier, which was applied for validation of three cohorts, and which achieved accuracy in diagnosing TCMR, AMR, and IFTA.…”

“…Other studies have shown little or no effect of complement deposition in vivo on AMR-related transcripts [2 & ]. Bouchet et al [8 ] studied follow-up biopsies of patients treated for CAMR and observed the disappearance and progression of Banff lesions (mi, C4d) and DSA in association with CAMR. The biopsy lesion patterns fell into three groups: disappearance (30%), persistence (47%), and progression (23%).…”

Humans versus machines: Who learns faster? Editorial regarding: “Machine learning-supported interpretation of kidney graft elementary lesions in combination with clinical data”

“…In this issue, Labriffe et al 3 applied a machine learning–supported interpretation of kidney allograft pathology lesions enhanced by combining it with clinical data. The machine learning took place in a large cohort ( n = 935) of biopsies, which were read by an expert panel of pathologist following the most recent Banff rules and further enhanced by consensus input from experienced transplant physicians considering the clinical context, together generating the best as it can get reference diagnosis for each biopsy.…”

“…Over time, the Banff rules increased in complexity to a point where humans became unable or unwilling to follow the rules. 2 In this issue, Labriffe et al 3…”