Vivien Holzwarth Correa

Vivien Holzwarth Correa

Bachelor's Thesis

Classification and assessment of rheumatoid arthritis from motion capture data using machine learning


Sophie Fleischmann (M. Sc.), Misha Sadeghi ( M. Sc.), Birte Coppers ( M. Sc.) Prof. Dr.-Ing. habil. Sigrid Leyendecker ,PD Dr. Anna-Maria Liphardt Robert Richer ( M. Sc.) Prof. Dr. Björn Eskofier


06 / 2023 – 11 / 2023


Rheumatoid Arthritis (RA) is an inflammatory rheumatic disease that leads to erosive joint damage and functional impairment [1]. This disease predominantly affects those joints of the body that are lined with a specialized tissue responsible for maintaining the nutrition and lubrication of the joint, such as the small joints of the hands. Besides joint damage, it has also become evident that people living with RA have a greater risk of cardiac failure, atherosclerosis, and osteoporosis [2]. Additionally, the flares of this disease can be severe, which puts RA patients at a higher risk of depression and anxiety [3]. On account of being a chronic disease that many adults develop around the mid stages of their lives, the diagnosis and especially the technique of monitoring RA should be improved and further researched. Identifying RA when it first presents itself and receiving treatment at an early stage can alter the course of the disease. The formation of joint erosions could be prevented, and the progression of the disease can significantly be slowed down. Close monitoring of the disease activity is furthermore needed to secure a timely and good medical adaptation and prevent flares/phases of high disease activity.[1]. Typically, rheumatoid arthritis is assessed by using the clinical findings from the patient’s medical history and examinations to conduct appropriate laboratory tests to confirm RA. Sometimes, diagnosis of RA may be possible based on clinical grounds alone, yet these parameters might not be sensitive enough for a precise diagnosis and monitoring of the disease. It would thus be beneficial to identify and incorporate additional objective functional biomarkers for the disease. Given that RA primarily affects the motion range of the patients, using marker-based motion capture to analyze movement restrictions in rheumatoid arthritis could offer a more objective approach to finding functional biomarkers, potentially improving the precision of diagnosis and disease monitoring beyond the limitations of clinical parameters alone. While motion capture data is typically complex and high-dimensional, machine learning can help to identify patterns and underlying relationships in the data that cannot be captured with traditional statistical tools. The use of marker-based motion capture and machine learning for the analysis of functional impairment in RA patients has not been extensively researched yet and the research gap is still large. The purpose of this thesis is to apply machine learning to identify movement restrictions and patterns associated with RA that have the potential to serve as functional biomarkers of the disease. Specifically, the goal is to investigate if we can distinguish RA patients from healthy controls based on movement data measured with optical motion capture, and if we can further predict the progress of the disease. Ultimately, we aim to evaluate the importance of different features in the classification process, thereby identifying relevant functional markers for characterizing RA. To do so, a binary classification pipeline will be implemented that differentiates between patients and non-patients. In addition, a multiclass classification algorithm will be used to predict different clinical patient scores related to disease severity. The input for the models will be features computed from motion capture and potentially EMG data of different hand motion tasks from a pilot study from 2021 [6]. We focus on hand movements, as the hand joints are significantly impacted in patients with rheumatoid arthritis. For example, RA leads to a reduction in hand grip strength and compromises the ability to curl the hand [5]. Feature selection algorithms will be part of the pipelines to select the most promising variables for the classification.


[1] Heidari B. Rheumatoid Arthritis: Early diagnosis and treatment outcomes. Caspian J Intern Med. 2011 Winter;2(1):161-70. PMID: 24024009 ; PMCID: PMC3766928.

[2] London: Royal College of Physicians Rheumatoid Arthritis: National clinical guideline for management and treatment in adults National Collaborating Centre for Chronic Conditions, February 2009

[3] Zhang, C. Flare‑up of cytokines in rheumatoid arthritis and their role in triggering depression: Shared common function and their possible applications in treatment (Review). Biomedical Reports, 14, 16., 2021

[4] Khan MH, Zöller M, Farid MS, Grzegorzek M. sl Marker-Based Movement Analysis of Human Body Parts in Therapeutic Procedure. Sensors (Basel). 2020 Jun 10;20(11):3312. doi: 10.3390/s20113312 . PMID: 32532113 ; PMCID: PMC7313697.

[5] Mohammed RH, Bhutta BS. Hand and Wrist Rheumatoid Arthritis [Updated 2023 Mar 11]. In: StatPearls [Internet]. Treasure Island (FL): StatPearls Publishing; 2023 Jan-

[6] Phutane, U.; Liphardt, A.-M.; Bräunig, J.; Penner, J.; Klebl, M.; Tascilar, K.; Vossiek, M.; Kleyer, A.; Schett, G.; Leyendecker, S.Evaluation of Optical and Radar Based Motion Capturing Technologies for Characterizing Hand Movement in Rheumatoid Arthritis — A Pilot Study Sensors 2021, 21, 1208. s21041208