Kidney Week 2025 Early Program - Glomerular Diseases: 2025 Update-Exploring the Impact of AI on Glomerular Diseases

Video Transcription

Video Summary

Dr. Lola Mariani, a clinician, reviews key AI and machine-learning concepts and how they apply to kidney disease research and care. She distinguishes learning “approaches” (unsupervised, supervised, deep learning, reinforcement learning) from clinical “tasks” (subgroup discovery, outcome prediction, object identification). She highlights pitfalls: nonrepresentative data, unstable clustering, missingness/outliers, poor or subjective labels, limited interpretability of neural networks, and challenges applying reinforcement learning and defining clinical “rewards.” She also explains generative AI, large language models, and foundational models, noting risks like bias and hallucinated citations. <br /><br />Mariani tours applications in nephrology, especially digital pathology: automated biopsy quality control (HistoQC), segmentation of kidney structures, and quantifying subtle “pathomic” features linked to outcomes. Emerging models combine pathology, pathomics, and clinical data to predict steroid benefit and assist diagnosis from EM images. She notes ML for risk prediction (IgA nephropathy, lupus flares) and for drug development, including mechanism inference, molecule generation, and simulating clinical trials using LLM-structured EHR data. She emphasizes privacy, rigorous validation, and proving real patient benefit.

Asset Subtitle

Laura Mariani

Meta Tag

Module GLOM

Speaker Laura Mariani

Keywords

AI in nephrology

machine learning approaches (supervised, unsupervised, deep learning, reinforcement learning)

digital pathology and kidney biopsy analysis

pathomics and outcome prediction

generative AI and large language models in healthcare

data quality, bias, interpretability, and clinical validation