Diabetes, Metabolic Syndrome and Obesity (Jul 2025)
Utilizing Podocyte Foot Process Morphology for the Identification of Diabetic Nephropathy with or without Minimal Change Disease: Establishment of an Artificial Intelligence-Assisted Diagnostic Model
Abstract
Xiangmeng Li,1– 4 Peihua Zhang,5 Shimin Jiang,1 Shunlai Shang,1 Jiao Zhang,6 Jinyu Liu,7 Chenchen Li,3,4 Yan Gao,3,4 Haisong Zhang,3,4 Wenge Li1,2,6,7 1Department of Nephrology, China-Japan Friendship Hospital, Beijing, People’s Republic of China; 2China-Japan Friendship Hospital, Chinese Academy of Medical Sciences & Peking Union Medical College, Beijing, People’s Republic of China; 3Key Laboratory of Bone Metabolism and Physiology in Chronic Kidney Disease of Hebei Province, Affiliated Hospital of Hebei University, Baoding, Hebei, People’s Republic of China; 4Department of Nephrology, Affiliated Hospital of Hebei University, Baoding, Hebei, People’s Republic of China; 5School of Big Data Science, Hebei Finance University, Baoding, Hebei, People’s Republic of China; 6Graduate School, Beijing University of Chinese Medicine, Beijing, People’s Republic of China; 7Graduate School, Capital Medical University, Beijing, People’s Republic of ChinaCorrespondence: Wenge Li, Department of Nephrology, China-Japan Friendship Hospital, Beijing, People’s Republic of China, Email [email protected] Haisong Zhang, Department of Nephrology, Affiliated Hospital of Hebei University, Baoding, Hebei, People’s Republic of China, Email [email protected]: This study aimed to differentiate whether diabetic nephropathy (DN) is complicated by minimal change disease (MCD) through the differences in podocyte foot process morphology, and subsequently establish an Artificial Intelligence-Assisted (AI-assisted) Diagnostic Model through machine learning of renal tissue electron microscopy images.Methods: Patients diagnosed with DN with nephrotic syndrome and treated in our hospital from January 2014 to December 2023 were selected. Patients were divided into the DN group and the DN with MCD group (DN+MCD group). Podocyte morphology’s diagnostic value was assessed by measuring foot process width and quantifying slit diaphragm changes via Nephrin immunohistochemical staining. This study pioneers developing a machine learning-powered diagnostic model based on renal electron microscopy imaging to differentiate DN cases with or without concurrent MCD.Results: In 51 patients, DN+MCD patients exhibited wider podocyte foot processes and reduced Nephrin expression compared to DN. A total of 622 electron microscopy images were used for model establishment and internal validation, while 225 electron microscopy images were used for external validation. A model based on Mobilenetv2 was successfully established, achieving a maximum accuracy of 93.3% in differentiating whether DN is complicated by MCD using a single image. When at least 11 random images were input, stable reports were obtained with an accuracy of 98%. External validation showed that the model had good sensitivity and specificity in differentiating whether DN is complicated by MCD (100%, 83.33%).Conclusion: Podocyte foot process morphology has diagnostic value in differentiating whether DN is complicated by MCD. Our AI model addresses the unmet clinical need for reliable differentiation between DN with and without concurrent MCD. Additionally, it establishes a foundational framework for AI-powered analysis of renal imaging data to improve disease diagnosis and prognosis prediction.Keywords: diabetic nephropathy, minimal change disease, podocyte, foot process width, slit diaphragm, machine learning
