Model Development and Application Based on Renal Pathology Images

N/AActive Not RecruitingNCT07816874

Union Hospital, Tongji Medical College, Huazhong University of Science and Technology8,000 enrolled

Overview

This single-center retrospective observational cohort study will utilize digital renal pathology images along with corresponding clinical and laboratory data from patients who underwent kidney biopsy at Union Hospital, Tongji Medical College, Huazhong University of Science and Technology, between January 1, 2019, and March 31, 2026. Deep learning and image analysis techniques will be employed to develop and validate models for the pathological diagnosis and classification of kidney diseases. The study will also assess model performance in evaluating disease activity and chronicity, as well as investigate the association between image-derived features and long-term renal outcomes. Diagnoses will be established by consensus among three senior renal pathologists, who will serve as the reference standard. A minimum of 8,000 eligible cases are planned for inclusion, which will be divided into training, validation, and independent test sets.

Study Type

OBSERVATIONAL

Enrollment

8,000

Conditions

Kidney Diseases, Chronic Glomerular Diseases Tubulointerstitial Kidney Diseases IgA Nephropathy (IgAN)Membranous Nephropathy

Eligibility

Sex: ALL

Medical Language ↔ Plain English

Inclusion Criteria: 1. Patients with kidney disease confirmed by kidney biopsy. 2. Availability of complete digital renal pathology images, including light microscopy, immunofluorescence microscopy, and/or electron microscopy images, as required for the relevant diagnostic task. 3. Availability of the clinical and laboratory data required for the planned analyses. Exclusion Criteria: 1. Renal pathology images of insufficient quality for analysis. 2. Missing key clinical data required for the planned analyses. 3. Kidney allograft biopsy specimens.

Outcomes

Primary Outcomes

Diagnostic classification accuracy of the deep learning model, measured as the percentage of correctly classified renal pathology cases

Diagnostic classification accuracy will be assessed in the independent test set by comparing the renal disease classification predicted by the prespecified deep learning model with the consensus diagnosis established by three senior renal pathologists. Accuracy will be calculated as the number of correctly classified cases divided by the total number of evaluable cases and reported as a percentage (%). The measurement tool will be the prespecified deep learning model evaluated against the expert consensus reference standard.

Time frame: At completion of independent test-set evaluation using retrospective data collected from January 1, 2019, through March 31, 2026.

Secondary Outcomes

Agreement between deep learning model predictions and expert consensus diagnoses, measured by the Cohen kappa coefficient

Agreement between the renal disease classifications generated by the prespecified deep learning model and the consensus diagnoses established by three senior renal pathologists will be assessed using the Cohen kappa coefficient. The measurement tool will be a prespecified categorical agreement analysis comparing model-predicted classifications with the expert consensus reference standard. The Cohen kappa coefficient is a unitless measure, with higher values indicating greater agreement beyond chance.