Photo Credit: gabrielhrech
The following is a summary of “Galileo—an Artificial Intelligence tool for evaluating pre-implantation kidney biopsies,” published in the October 2024 issue of Nephrology by Eccher et al.
Pre-transplant kidney biopsy interpretation is difficult due to a lack of renal pathology experts. The “Galileo” artificial intelligence tool supports pathologists in assessing kidney donor biopsies.
Researchers conducted a retrospective study to evaluate the effectiveness of the “Galileo” artificial intelligence tool in assisting pathologists with pre-implantation kidney biopsy interpretations.
They collected a multicenter cohort of whole slide images from core-needle and wedge biopsies of the kidney. A deep learning algorithm was trained on the Aiforia Create platform to identify key findings in the pre-implantation setting normal glomeruli, globally sclerosed glomeruli, ischemic glomeruli, arterioles, and arteries and validated on an external dataset by three independent pathologists to assess algorithm performance.
The results showed that Galileo achieved a precision of 81.96%, sensitivity of 94.39%, F1 score of 87.74%, and total area error of 2.81% in the training set, and 74.05%, 71.03%, 72.5%, and 2% in the validation set, respectively. It was significantly faster than pathologists, requiring only 2 minutes overall in the validation phase compared to 25, 22, and 31 minutes by 3 separate human readers (P<0.001). Galileo-assisted detection of renal structures and quantitative information was integrated into the final report.
Investigators concluded that the Galileo AI-assisted tool improved the speed of pre-implantation kidney biopsy interpretation and reduced inter-observer variability, potentially serving as a foundation for future enhancements related to graft survival.
Source: link.springer.com/article/10.1007/s40620-024-02094-4
