AI-based computer model accurately analyzes kidney biopsy images
Boston University School of Medicine has developed computer models based on artificial intelligence that significantly improve the analysis of routine kidney biopsy images.
BU researchers, who conducted a proof-of-principle study on kidney biopsy sections, contend that their AI-based models have both diagnostic and prognostic applications and could lead to the development of software for diagnosing kidney disease as well as predicting kidney survival.
In the study, images processed from renal biopsy samples were collected on 171 patients treated at the Boston Medical Center and were analyzed by convolutional neural networks (CNN) models and nephropathologists, who specialize in the analysis of kidney biopsy images.
“With respect to kidney disease, biopsy is one of the gold standard procedures,” says Vijaya Kolachalama, lead author and assistant professor of medicine at Boston University School of Medicine. “Most of the clinical decisions today are made based on information that nephropathologists can see from the biopsy.”
However, in the study—published last week in the journal Kidney International Reports—CNN models outperformed human kidney analysis across six disease classification tasks. In particular, the AI-based model predicted the chronic kidney disease stage more accurately than the nephropathologists.
"When implemented in the clinical setting, our work will allow pathologists to see things early and obtain insights that were not previously available,” claims Kolachalama, who notes that the digitization of biopsies is not yet a common practice but is growing.
Kolachalama believes that with a shortage of nephropathologists the application of CNN to the analysis of routinely obtained kidney biopsies can help to alleviate this lack of manpower with the ability of AI-based models to quantify the extent of renal damage and predict the life remaining in the kidney.
"While the trained eyes of expert pathologists are able to gauge the severity of disease and detect nuances of kidney damage with remarkable accuracy, such expertise is not available in all locations, especially at a global level,” he notes. “If healthcare providers around the world can have the ability to classify kidney biopsy images with the accuracy of a nephropathologist right at the point-of-care, then this can significantly impact renal practice. In essence, our model has the potential to act as a surrogate nephropathologist, especially in resource-limited settings.”
Researchers will continue to refine the accuracy of the AI-based models. This research is personal for Kolachalama, whose father has end stage kidney disease and whose mother had a minor form of kidney disease. Ultimately, he says the goal is to build software that can provide this powerful tool to provider organizations “anywhere, at any location” and not just large academic medical centers.
A separate editorial published in Kidney International Reports along with the BU study contends that in order to bring these CNN models into clinical practice “researchers and clinicians should not be solely preoccupied with speed and accuracy, but should assess how these methods integrate into the medical setting.”