Advancing the application of the analytical renal pathology system in allograft IgA nephropathy patients

Figures & data

Figure 1. Workflow of our study. At stage I, the analytic renal pathology system (ARPS) was established. A dataset consisting of 41,000 patches obtained from 400 native immunoglobulin A nephropathy (IgAN) patients was utilized, with the patches carefully labeled for the location and identification of glomeruli lesions (GS, SS, C and NOA). Moreover, for the identification of intrinsic glomerular cells, we selected and labeled 460 images of glomeruli (about 70 000 cells, including M, E and P) from NOA samples. The ARPS system exhibited exceptional performance in accurately categorizing glomerular lesion types and intrinsic glomerular cells (13). Stage II is allograft tissue compartment recognition, where periodic acid-Schiff (PAS) slides of allograft IgAN patients were scanned and then uploaded. After that, the ARPS identified the glomerular lesions and internal glomerular structures using U-Net, DenseNet, LSTM-GCNet and 2D V-Net. Stage III is the whole-slide image (WSI) clinical investigation, where we evaluated the reliability of the ARPS relative to ground truth (evaluated by a pathologist) in identifying glomerular lesions and internal glomerular structures. Meanwhile, the glomerular area, mesangial area, glomerular lesions (SS, GS, C and NOA) and intrinsic glomerular cells (M, E and P) automatically identified by the ARPS were further examined through association with clinical data and post-transplantation graft survival.

Table 1. Demographic And clinical characteristics of patients with allograft IgA nephropathy (IgAN) from internal validation center.

Items	Values
Age at transplantation, years	30.18 ± 8.18
Native biopsy
IgAN, n (%)	8 (14.29)
MPGN, n (%)	1 (1.79)
Sclerosing glomerulonephritis, n (%)	1 (1.79)
Donor biopsy
IgAN, n (%)	2 (3.57)
Undefined, n (%)	6 (10.71)
At graft biopsy
Age, years	32 (28, 39)
Gender (male), n (%)	43(76.79)
BMI, kg/m^2	22.45 ± 3.87
Graft age at allograft IgAN diagnosis, months	48.43 ± 29.52
Serum creatinine, mg/dL	1.44 (1.18, 2.00)
eGFR, mL/min/1.73 m^2	60.85 ± 26.72
Blood urea nitrogen, mg/dL	21.42 (16.30, 28.49)
Serum uric acid, umol/L	389.88 ± 85.05
Serum albumin, g/L	39.60 (36.60, 42.30)
Hemoglobin, g/L	125.71 ± 23.87
Total cholesterol, mmol/L	4.68 (3.87, 5.47)
Triglyceride, mmol/L	1.35 (0.93, 1.89)
Urine protein, g/24 h	0.76 (0.43, 1.86)
Microscopic hematuria, /uL	32.80 (8.73, 133.08)
Oxford Classification, n (%)
M1	17 (30.36)
E1	13 (23.21)
S1	31 (55.36)
T1/T2	8 (14.29)/5 (8.93)
C1/C2	19 (33.93)/3 (5.36)

Values for continuous variables were expressed as mean ± SD or median (interquartile range), and categorical variables were expressed as number (percentage). MPGN, membranoproliferative glomerulonephritis; BMI, body mass index; eGFR, estimated glomerular filtration rate; M, mesangial hypercellularity; E, endocapillary hypercellularity; S, segmental glomerular sclerosis; T, interstitial fibrosis and tubular atrophy; C, crescent.

Figure 2. Identification of glomerular lesion types and intraglomerular structures in periodic acid–Schiff (PAS) slides from allograft IgAN patients by the analytic renal pathology system (ARPS). (A) A section of a whole slide image (WSI) from allograft IgAN patients. (B) Segmentation and location of glomerular lesion types in the WSI by the ARPS. Each glomerulus is categorized into global glomerular sclerosis (green), segmental glomerular sclerosis (yellow), crescents (red), and none of the above (blue). [

], the glomerular lesion was identified incorrectly. (C) A glomerulus (none of the above) in a WSI. (D) Intrinsic glomerular cell segmentation and identification, including mesangial cells (blue), endothelial cells (yellow), podocytes (green), and glomerulus boundary. [

], the intrinsic glomerular cell was not identified and located; [

], the intrinsic glomerular cell was identified incorrectly; [

], other internal glomerular structure was misidentified as intrinsic glomerular cell. (E) Location of mesangial region and mesangial cells.

Figure 3. Metrics of the analytic renal pathology system (ARPS) on identifying glomerular lesions and intrinsic glomerular cells in allograft IgAN patients from the internal and external validation center. (A-B) Metrics of the ARPS on identifying glomerular lesions. (C-D) Metrics of the ARPS on identifying intrinsic glomerular cells. AUC-ROC, Area Under the Curve-Receiver Operating Characteristic; GS, global glomerular sclerosis; SS, segmental glomerular sclerosis; C, crescents; NOA, none of the above; M, mesangial cells; E, endothelial cells; P, podocytes. (E) Kappa agreement testing between the ARPS and a pathologist.

Table 2. ICCs Of glomerular lesions in different cohorts.

Glomerular lesions	ARPS-pathologist in allograft IgAN	Different pathologists in allograft IgAN	Different pathologists in native IgAN^20
GS, %	0.875	0.787	0.90
SS, %	0.651	0.507	0.46
C, %	0.547	0.366	0.32–0.66
NOA, %	0.929	0.874	–

ICC, intraclass correlation coefficient, which reflects the reliability and reproducibility of measurements or ratings between the responses of two subjects from the same cluster. Conventionally, an ICC of 0.80-1.00 is outstanding inter-rater reliability, 0.60–0.79 is substantial, 0.40–0.59 is moderate, and 0-0.39 is poor (19, 30). GS, global glomerular sclerosis; SS, segmental glomerular sclerosis; C, crescents; NOA, none of the above.

Figure 4. Correlation of mesangial digital features with Oxford mesangial hypercellularity score. M score, mesangial hypercellularity score from the Oxford classification of IgA nephropathy (19, 20). Cases in which more than half the glomeruli contain more than three cells in a mesangial area is categorized as M1, and conversely, were categorized as M0. P values and correlation coefficients were calculated from Spearman’s correlation test.

Figure 5. Kaplan–Meier curves of intrinsic glomerular cells identified by the analytic renal pathology system (ARPS) in high-level and low-level groups. E-max, maximum of endothelial cells; E-ratio, ratio of endothelial cells; P-max, maximum of podocytes; P-density, density of podocytes. P values were calculated by log-rank tests.

Table 3. Proportional Cox regression model assessing clinicopathological predictors of allograft failure.

Characteristics	Univariate		Multivariate
	HR (95% CI)	p value	HR (95% CI)	p value
Maximum of mesangial cells	0.986 (0.959–1.014)	.319
Maximum of endothelial cells	0.954 (0.918–0.990)	.014	–	–
Maximum of podocytes	0.887 (0.816–0.964)	.005	0.904 (0.825–0.990)	.030
Density of mesangial cells	1.066 (0.938–1.211)	.327
Density of endothelial cells	0.805 (0.657–0.985)	.036	–	–
Density of podocytes	0.942 (0.795–1.117)	.492
Ratio of mesangial cells	1.095 (1.011–1.186)	.026	–	–
Ratio of endothelial cells	0.870 (0.768–0.985)	.028	–	–
Ratio of podocytes	0.964 (0.873–1.065)	.468
Mesangial area ratio	1.152 (1.024–1.296)	.019	–	–
Graft age at allograft IgAN diagnosis	1.001 (0.985–1.017)	.937	–	–
Gender	2.623 (0.925–7.435)	.070	–	–
Age at graft biopsy	1.001 (0.947–1.057)	.984	–	–
eGFR	0.957 (0.935–0.979)	<.001	–	–
Serum creatinine	1.827 (1.353–2.468)	<.001	1.652 (1.199–2.275)	.002
Blood urea nitrogen	1.032 (1.005–1.060)	.022	–	–

eGFR, estimated glomerular filtration rate; HR, hazard ratio; CI, confidence interval.

Zeng C, Nan Y, Xu F, et al. Identification of glomerular lesions and intrinsic glomerular cell types in kidney diseases via deep learning. J Pathol. 2020;252(1):53–64. doi: 10.1002/path.5491.

 PubMed Web of Science ®Google Scholar

Roberts ISD, Cook HT, Troyanov S, Working Group of the International Ig ANN, the Renal Pathology S., et al. The oxford classification of IgA nephropathy: pathology definitions, correlations, and reproducibility. Kidney Int. 2009;76(5):546–556. doi: 10.1038/ki.2009.168.

 PubMed Web of Science ®Google Scholar

Zeng CH, Le W, Ni Z, et al. A multicenter application and evaluation of the oxford classification of IgA nephropathy in adult chinese patients. Am J Kidney Dis. 2012;60(5):812–820. doi: 10.1053/j.ajkd.2012.06.011.

 PubMed Web of Science ®Google Scholar

Cattran DC, Coppo R, Cook HT, Working Group of the International Ig ANN, the Renal Pathology S., et al. The oxford classification of IgA nephropathy: rationale, clinicopathological correlations, and classification. Kidney Int. 2009;76(5):534–545. doi: 10.1038/ki.2009.243.

 PubMed Web of Science ®Google Scholar

Data availability statement

The data underlying this article will be shared on reasonable request to the corresponding author.