Reply to: “On the Limitations of Computer Vision Models for Movement Disorders”

We thank Drs. Coebergh and Lees for their interest, appreciate the active discussion regarding our review on video-based models for movement disorders, and welcome these insights into their implications for the movement disorders community. We strongly agree that these models should not be used for clinical decision-making without the involvement of human experts. Their appropriate role is as assistive tools, providing quantitative measurements that support, but do not replace, a clinician's integral assessment. Movement disorders extend beyond what is visually observable, and non-motor symptoms are not detected by the approaches covered in our review. Nevertheless, we foresee that data-driven decision support tools will provide unique advantages to clinical work in the future. The literature included in our review suggests that computer models are increasingly able to identify disorder-related movements, but research on complex cases, including functional movement disorders, remains scarce. Some works suggest that models can distinguish multiple movement types at once,1 but these are not validated for medical use, nor for movement disorders. Part of the challenge is data related: even for single phenotypes, data are often constrained. Critically, these data require meticulous labeling and validation by medical experts. For complex cases, this is a severe limitation, as clinicians may not have enough agreement on a diagnosis2 to provide a solid ground truth for a model. To provide reliable clinical support, research must prioritize explainable models over “black box” models. This will allow clinicians to understand how a model is arriving at its conclusion and how it connects to an observed symptom. This is not yet the case in current literature, but it is an attainable goal3 for future works in the field. Nevertheless, there are reasons for optimism. Data-driven techniques enable contact-free measurement and can accurately identify features4 that strongly correlate to a diagnosis, such as the frequency of movements.5 This can provide new insights to make a decision during patient examinations in clinical work, or aid in uncovering unknown aspects of a disorder's pathophysiology in medical research. As models improve, they can provide less experienced clinicians with support for their hypothesized movement disorder diagnosis and assist with their decision to initiate treatment or, for complex cases, suggest the involvement of a movement disorders specialist. Non-motor symptoms observed by clinicians can be used for analysis through multimodal algorithms, incorporating patient histories into a model's reasoning.6 Accessible and broadly available technologies such as these may help close the distance to places where clinical experts are most needed by lowering the time and resources needed to obtain thorough measurements. We highlight the promise of these tools as complementary ways to observe a patient, not as a replacement for clinical expertise. We believe continued research is justified given the complexity of movement disorders. This will allow the community to better understand where these models break and to improve their usability so that they may become a reliable addition to our clinical toolkits. (1) Research project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript: A. Writing of the first draft, B. Review and Critique. Rafael Martínez-García-Peña: 1A, 1B, 1C, 2A, 2B, 3A. Lisette H. Koens: 1A, 1B, 2A, 2B, 3B. George Azzopardi: 1A, 1B, 1C, 2A, 2B, 2C, 3B. Marina A.J. Tijssen: 1A, 1B, 1C, 2C, 3B. The authors declare that financial support was received for the authorship of the original article (NWO Grant File 20429). Disclosure of all funding since last year: All authors are employed as stated in the affiliations. R.M.-G.-P. and L.H.K. report no other sources of funding. G.A. reports a grant from CogniGron. M.A.J.T. reports grants from the Netherlands Organisation for Health Research and Development ZonMW Topsubsidie (91218013), the European Fund for Regional Development from the European Union (01492947) and the province of Friesland, the Dystonia Medical Research Foundation, and the Dystonie Wetenschapsfonds, and unrestricted grants from Actelion and Merz. Author disclosures are available in the Supporting Information. Data S1. Coi_disclosure_letterreply. Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.