Machine learning for classifying the early stage of Osteoarthritis based on biological data

Abstract

Osteoarthritis, or OA, is a chronic joint disease and the most common form of arthritis. It is a very common disease in human athletes, but also the most common reason for lameness and poor performance in animal athletes, such as racehorses. The traditional standard for diagnosing OA is by radiographic measurements. Unfortunately, clinically recognizable changes do not appear until the chronic destruction of the articular cartilage has progressed too far and the disease is irreversible. In order to diagnose the disease earlier, the focus has been shifted from imaging biomarkers to biological biomarkers. Several promising biological biomarkers have been found by researchers at SLU and Sahlgrenska, each representing a different stage of the destruction process. One specific biomarker has shown to increase in both blood and synovial fluid in horses with acute lameness, corresponding to an early stage of OA. If this early OA could be identified, it would be possible to intervene in time and the chronic and painful destruction of the joint tissues could be prevented, which could greatly improve the equine welfare. The aim of this thesis was to investigate different machine learning approaches in order to find a promising method to be used in a decision support system for practitioners. The future system should be able to help diagnose OA, and specifically identify the different progression stages of structural changes in the joint, based on biological data. A Random Forest Classifier was developed along with a Spectral Clustering Algorithm, which was trained and evaluated on datasets with samples from both synovial fluid and serum. The results indicate some promise for the future decision support system, which will have to be evaluated further once more data is collected and the biomarkers for the remaining progression stages are added in the mix.