Chronic Changes on Kidney Histology by a Multiclass Artificial Intelligence Model.

Area of Science:

• Digital pathology
• Artificial intelligence in medicine
• Nephrology

Background:

• Assessing chronic changes in kidney histology is crucial for predicting kidney disease outcomes.
• Current methods relying on human interpretation are subjective and have limited accuracy.
• Automated quantification of histological features can improve diagnostic consistency and prognostic value.

Purpose of the Study:

• To develop and validate an artificial intelligence (AI) model for segmenting and quantifying chronic changes in kidney histology whole slide images (WSIs).
• To assess the association between AI-derived chronic kidney disease (CKD) measures and clinical outcomes in kidney donors and patients with renal tumors.

Main Methods:

• An AI model was trained using 20,509 annotations across 20 classes on kidney tissue WSIs.
• The AI model's performance was validated against human annotations, comparing agreement levels.
• The AI model quantified chronic changes, including nephron size and nephrosclerosis markers, in 1426 donors and 1699 tumor patients.

Main Results:

• The AI model demonstrated comparable agreement to human pairs in most detections, with lower agreement for arteriolar hyalinosis (AH).
• AI-derived chronic changes correlated with reduced glomerular filtration rate (GFR) post-donation and kidney failure post-nephrectomy.
• A chronicity score derived from AI detections showed strong prognostic capability for kidney failure (C-statistic = 0.819).

Conclusions:

• A multiclass AI model effectively automates the quantification of chronic kidney disease changes from WSIs.
• AI-based histological assessment offers a reliable tool for predicting CKD progression and outcomes.
• This technology has the potential to enhance diagnostic accuracy and patient management in nephrology.