关键词: 人工智能, 病理, 原发性膜性肾病, 肾小球病变类型, 固有细胞

Abstract: Ｏｂｊｅｃｔｉｖｅ：To explore the application of artificial intelligence-based analytic renal pathology system (ARPS) in patients with primary membranous nephropathy (PMN).
Ｍｅｔｈｏｄｏｌｏｇｙ：Patients with biopsy-proven PMN in our center from January 2018 to December 2019 were enrolled. Their clinical and pathological data were collected. ARPS was applied to identify glomeruli lesions including global glomerular sclerosis (GS), segmental glomerular sclerosis (SS), crescents (C), and none of the above (NOA), and then quantify intraglomerular features, including intrinsic glomerular cells (M: mesangial cells; E: endothelial cells; P: podocytes) and glomerular area. The performance of ARPS was evaluated. The relationships between ARPSbased intraglomerular features and prognosis were further analyzed.
Ｒｅｓｕｌｔｓ：A total of 123 patients (65.0％ males) were included. The average age was 47.1±14.0 years at biopsy. For the identification of glomerular lesions, ARPS achieved the best effect in NOA with 0.967 F1-score; followed by GS with 0.811 F1-score and SS with 0.545 F1-score. The F1-scores of ARPS on identifying intrinsic cells were greater than 0.950. The glomerular area was larger in no response (NR) patients than that in partial remission (PR) and complete remission (CR) patients; the number, density and ratio of podocytes were higher in CR patients than that in PR and NR patients. The remission rate of urinary protein was higher in patients with higher podocytes number or density than that in patients with lower podocytes number or density. Renal phospholipase A2 receptor (PLA2R) and podocyte number were independent predictors of urinary protein remission.
Ｃｏｎｃｌｕｓｉｏｎ：ARPS performed well in the identification of glomerular lesions and intrinsic cell types in PMN. The ARPSbased intraglomerular characteristics were significantly correlated with prognosis in PMN patients.

Key words: artificial intelligence, renal pathology, primary membranous nephropathy, glomerular lesionsintrinsic cells