Machine Learning Changes the Dynamics of Kidney Disease Diagnosis

Kidney disease is following other disorders as a target for new and improved machine learning. As published in a recent edition of Science Daily, pathologists have often classified kidney diseases based on visual assessments of kidneys.

Machine learning, technically a branch of AI, has shown potential in increasing the accuracy for classifying these diseases. It was conceived with the idea that machines can not only process data, they can learn without human supervision. Machine learning includes learning from data, identifying patterns and making decisions.

Two Studies on the subject were published by (JASN) Journal of the American Society of Nephrology.

The first study involved two doctors at the Jacobs School of Medicine who led their team in developing a computational formula (algorithm) that functions entirely on its own to assess diabetic kidney disorder.

The formula has been developed to examine digital images of a kidney biopsy. Information is extracted on the glomeruli. These are small capillary blood vessels in the kidney where waste is filtered and forms urine. Diabetes causes progressive damage and scarring to the glomeruli.

Every kidney biopsy involves ten to twenty individual glomeruli. The algorithm finds their sub-component in the digital images and records measurements for each.

Just as a doctor scans a patient’s biopsy, the algorithm examines the structure of each of the glomeruli it has recorded and incorporates that information into its analysis.

The samples from fifty-four diabetics who have kidney disease were digitally classified by the researchers. They were found to conform with the classifications presented by three pathologists.

The Second Study

A second JASN article described another study that was conducted at the Radboud University Medical Center in the Netherlands. The team used machine learning in their examination of biopsies that were taken from patients who had kidney transplants. In addition to examining the glomeruli, this study assessed various classes of kidney tissue.

Convolutional Neural Network (CNN)

CNN is a deep learning model in a class of neural networks. CNN may be applied to tissues from biopsies and samples from nephrectomies (kidney removal). The model is effective in analyzing diseased and healthy tissues. Its methods of classification conform to those used in standard methods.

Artificial Intelligence (AI)

The researchers applied AI for accuracy in analyzing tissue from kidney transplantation. Dr. van der Laak, the co-leader of the study, explained that the use of AI will yield more accurate data. It will also improve diagnosis and organ survival for patients who had undergone transplants.

The doctor was pleased to announce that CNN’s performance surpassed their expectations. However, the team included eight additional classes of tissue. In this instance, the network did not outperform the human observer.

The team was to determine whether certain essential structures (renal tubules) were in an atrophic state (wasting away). The team is currently striving for improved accuracy in this classification.

At this point in time, machine learning has only been applied to the kidneys and limited to analyzing one structure.

Researchers working with machine learning believe that in the future they can extract more information from kidney tissue to support assessment for a kidney transplant. The network provides an opportunity for deep learning applications in standard diagnostics.

Rose Duesterwald          March 19, 2020

Rose Duesterwald

Rose Duesterwald

Rose became acquainted with Patient Worthy after her husband was diagnosed with Acute Myeloid Leukemia four years ago. He was treated with a methylating agent While he was being treated with a hypomethylating agent, Rose researched investigational drugs being developed to treat relapsed/refractory AML.

Share this post

Share on facebook
Share on google
Share on twitter
Share on linkedin
Share on pinterest
Share on print
Share on email
Close Menu