Scientists have used computational tools, including machine learning, to differentiate between subtypes of rheumatoid arthritis. In this study, which was reported in Nature Communications, the researchers created an algorithm that can analyze pathology slides from RA patients automatically. The study authors suggested that this could make the diagnosis and treatment of RA more efficient and personalized.
Advanced computational tools could also advance other aspects of medical science, they noted. This study has demonstrated that machine learning might transform how ther pathologies of various diseases are assessed, said senior study author Dr. Fei Wang, a Professor at Weill Cornell Medicine, among other appointments.
Scientists studying a variety of conditions such as cancer have been working to automate the analysis of pathology slides, which are used to examine patient specimens. Others are trying to expand this technology for use in the diagnosis of other conditions.
Arthritis is typically diagnosed by clinicians and pathologists who use a rubric to assess the characteristics of cells and tissues from patient biopsies. This is subjective work, so a diagnosis may depend on who is performing the analysis. But clinicians have to determine what subtype of RA a patient has to identify the best course of treatment.
"It's the analytical bottleneck of pathology research. It is very time-consuming and tedious," said first study author Dr. Richard Bell, a computational pathology analyst at the Hospital for Special Surgery (HSS), among other appointments.
In this study, the investigators trained a machine learning algorithm on mouse data, and optimized its conclusions so that it could distinguish between RA subtypes. Another set of data was used to validate the tool. The researchers gained some new treatment insights from this effort, which were used on their mouse models. There were reductions in cartilage degeneration in the treated mice within six weeks of starting the appropriate therapeutic, which had been indicated by the computational tool.
Next, the team utilized the tool to analyze patient samples. This showed that the algorithm could correctly and efficiently identify RA subtypes in human samples. Additional validation is now taking place.
The researchers are hoping to incorporate this tool into RA care soon. "It's the first step towards more personalized RA care," Bell said. "If you can build an algorithm that identifies a patient's subtype, you'll be able to get patients the treatments they need more quickly."
They are also working to create similar tools for diagnosing other conditions such as osteoarthritis and disc degeneration.
The method may also help scientists learn more about RA, if it can detect unusual characteristics in patient samples that technicians or pathologists might not find.
Sources: Weill Cornell Medicine, Nature Communications