Machine Learning Algorithms Promising in Kidney Diagnostics

Machine learning algorithms could aid with classification of kidney biopsy samples, according to two studies published online in the Journal of the American Society of Nephrology.

Brandon Ginley, from The State University of New York in Buffalo, and colleagues developed a digital pipeline to classify renal biopsies from patients with diabetic nephropathy (DN). Traditional image analysis was combined with modern machine learning to capture important structures. A set of digital features was defined that quantified the structural progression of DN. The researchers found that the digital classification agreed with a senior pathologist, with moderate Cohen's kappa (0.55). Two other renal pathologists agreed with the digital classification (κ = 0.68 and 0.48, respectively).

Meyke Hermsen, from Radboud University Medical Center in Nijmegen, Netherlands, and colleagues trained a convolutional neural network for multiclass segmentation of digitized kidney tissue sections using 40 whole-slide images of stained kidney transplant biopsies. The researchers found that the weighted mean Dice coefficients were 0.80 and 0.84, respectively, for all classes in 10 kidney transplant biopsies from the Radboud center and an external center. In both data sets, the best segmented class was "glomeruli" (Dice coefficients, 0.95 and 0.94, respectively) followed by "tubuli combined" and "interstitium." In nephrectomy samples, the network detected 92.7 percent of all glomeruli, with 10.4 percent false positives. The mean intraclass correlation coefficient for glomerular counting performed by pathologists versus the network was 0.94 in whole transplant biopsies.

"There are indications that the best solution may be augmented intelligence, with clinician and machine learning working together," write the authors of an accompanying editorial.

-----------------------------------------------------------------------
Editor's Note:

For any copyright disputes involving the content, 

please email: Julia.Zhang@ubmsinoexpo.com to delete.