Galileo-an Artificial Intelligence tool for evaluating pre-implantation kidney biopsies

Area of Science:

• Nephrology
• Digital Pathology
• Artificial Intelligence

Background:

• Pre-transplant kidney biopsy interpretation is complex and time-consuming.
• A shortage of renal pathology experts exacerbates interpretation challenges.
• Artificial intelligence (AI) offers a solution to assist pathologists in kidney donor biopsy assessment.

Purpose of the Study:

• To introduce the "Galileo" AI tool for assisting pathologists with pre-implantation kidney biopsies.
• To evaluate the performance of the AI tool in detecting key renal structures and findings.

Main Methods:

• A multicenter cohort of kidney biopsy whole slide images was used.
• A deep learning algorithm was trained to identify normal, globally sclerosed, and ischemic glomeruli, as well as arterioles and arteries.
• The AI model was validated by three independent pathologists on an external dataset.

Main Results:

• Galileo achieved high precision (81.96%) and sensitivity (94.39%) in the training set.
• In the validation set, Galileo demonstrated a precision of 74.05% and sensitivity of 71.03%.
• Galileo significantly reduced interpretation time to 2 minutes compared to 22-31 minutes for human pathologists (p < 0.001).

Conclusions:

• The Galileo AI tool shows promise in accelerating pre-transplant kidney biopsy interpretation.
• AI assistance can help reduce inter-observer variability among pathologists.
• This AI tool may serve as a foundation for future improvements impacting graft survival.