Artificial intelligence-assisted identification of glomerular pathological features in IgA nephropathy and Henoch-Sch?nlein purpura nephritis
Objective The study aims at evaluating the performance of the analytic renal pathology system(ARPS)based on deep learning in patients with Henoch-Schönleinpurpura nephritis(HSPN),and analyzing the relationship between the glomerular fea-tures quantified by the ARPS and clinical prognosis in patients with IgA nephropathy(IgAN)or HSPN.Methods The data of pa-tients diagnosed with IgAN(n=170)and HSPN(n=153)by renal biopsy in the National Clinical Experimental Center for Kidney Dis-eases from January 2018 to December 2019 were retrospectively analyzed.The precision,recall and F1 score were used to evaluate the recognition effect of ARPS in HSPN patients.ARPS was also applied to identify glomerular lesion types,such as glomerular scle-rosis(GS),segmental sclerosis(SS),crescents(C)and non-men-tioned glomerular lesions(NOA),and to identify the intrinsic cells in NOA glomeruli,including mesangial cells(M),endothelial cells(E)and podocytes(P).X-tile software was used to determine the optimal cutoff point of podocyte digital pathological features,and patients were divided into high podocyte value group(＞optimal cut-off point)and low podocyte value group(≤ optimal cutoff point).The clinical and pathological features of IgAN and HSPN patients were compared,and the relationship between digital pathological features and clinical prognosis was analyzed.Results ARPS has a good effect on identifying the lesion types of HSPN patients,among which the recognition effects of NO A and GS glomeruli(F1 scores of 0.95 and 0.90,respectively)are better than those of C and SS(F1 scores of 0.83 and 0.73,respectively).ARPS can accurately locate the glomerular intrinsic cells in patients with HSPN(F1 score＞0.91).The mesangial cell digital index of both groups of patients was positively correlated with the Oxford classification mesangial cell proliferation score(P＜0.01).The renal function stability rate of pa-tients in the high podocyte value group was significantly higher than that in the low podocyte value group(P＜0.05).Conclusion The ARPS can be applied to identify glomerular lesions and intrinsic cells in patients with HSPN.The mesangial cell digital index can indicate the degree of mesangial cell proliferation,and the podocyte digital index can assist in analyzing the prognosis of the two diseases.
deep learningHenoch-Schönlein purpura nephritisIgA nephrologypathology prognosis
NETL
NSTL
万方数据
