Machine Learning Effectively Characterizes Rheumatoid Arthritis Subtypes

Credit: Adobe Stock

A machine learning tool created by investigators at Weill Cornell Medicine can help identify subtypes of rheumatoid arthritis (RA) and may aid in preclinical and clinical research as a cost-effective phenotyping tool.¹

“Our tool automates the analysis of pathology slides, which may one day lead to more precise and efficient disease diagnosis and personalized treatment for RA,” Fei Wang, PhD, professor, population health sciences, and founding director, Institute of AI for Digital Health (AIDH), Department of Population Health Sciences, Weill Cornell Medicine, said in a statement.² “It shows that machine learning can potentially transform pathological assessment of many diseases.”

Wang and colleagues developed an automated multi-scale computational pathotyping (AMSCP) pipeline tool for human and mouse synovial tissue which was able to segment different tissue types to characterize tissue-level changes and classify cell types within each tissue compartment that assesses change across disease states.¹

After training, the tool was able to distinguish between pauci-immune, diffuse and lymphoid tissue phenotypes and between arthritis effector cells across healthy, mild disease, and severe disease cell types in mice in a manner sensitive to both subtle and dramatic tissue changes. The model was also validated in human cells from a human synovial biopsy data set from the Accelerating Medicines Partnership RA (AMP-RA) research consortium and was able to predict clinically relevant lymphoid cells in RA. The model has the potential to reduce the analytical bottleneck associated with histopathology assessments in both the clinical and preclinical settings.¹

“It’s the first step towards more personalized RA care,” investigator Richard Bell, PhD, MS, Postdoctoral Associate in Medicine, Weill Cornell Medical College, added.² “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.”

The model’s predictions were in line with established cell distributions within each phenotype, specifically that synovial fibroblast enrichment is found in the pauci-immune pathotype and some diffuse cases, while lymphocytes and plasma cells are found primarily in the lymphoid pathotype. Percent plasma cells had a high predictive capability for discriminating between diffuse and lymphoid cases (Area under a receiver operator curve = 0.82 ± 0.06, P <.0001. These data show that the model, or other similar models, can be used in clinically meaningful scenarios and potentially in clinical trials.¹

"By integrating pathology slides with clinical information, this tool demonstrates AI's growing impact in advancing personalized medicine," Rainu Kaushal, MD, MPH, senior associate dean for clinical research and chair, Department of Population Health Sciences, Weill Cornell Medicine, added.² "This research is particularly exciting as it opens new pathways for detection and treatment, making significant strides in how we understand and care for people with rheumatoid arthritis."

The investigators noted that a formal independent validation cohort needs to be developed in order to realize the full potential of the human models and understand unbiased performance. Another limitation is that multiclass cell typing was limited by staining methods. They also emphasized the potential of the model to aid in characterizing disease beyond inflammatory RA, although further research and quality control of the modeling is needed for a true impact.

REFERENCES

1. Bell RD, Brendel M, Konnaris MA, et al. Automated multi-scale computational pathotyping (AMSCP) of inflamed synovial tissue. Nat Comm. 2024(15):7503. doi: 10.1038/s41467-024-51012-6
2. Machine learning helps identify rheumatoid arthritis subtypes. News release. Weill Cornell Medicine. August 29, 2024. https://www.eurekalert.org/news-releases/1056185