Abstract Details

Title Machine Learning Analysis of Archimedes Spirals in Movement Disorders and Current Underuse in Ataxia Syndromes: A Systematic Review

Topic Movement Disorders

Presentation(s) P5 - Poster Session 5 (11:45 AM-12:45 PM)

Poster/Presentation
Number 16-006

Objective To evaluate the diagnostic utility of machine learning (ML) applied to Archimedes spirals for movement disorders and to identify its underuse in ataxia syndromes.

Background Various forms of cerebellar ataxia exist and remain challenging to diagnose due to its subtle and heterogeneous presentation. The Archimedes spiral is a simple, well-established tool for assessing tremor and coordination. Recent studies have applied ML to spiral drawings for tremor syndrome diagnosis and severity quantification; however, its application to cerebellar disorders is limited.

Design/Methods A systematic review was performed in PubMed, Web of Science, and Google Scholar to identify peer-reviewed, English-language studies utilizing ML models for pen-and-paper Archimedes spiral analysis. Book chapters and review papers were excluded. Extracted data included disease type, model architecture, and reported performance metrics (AUROC, accuracy, precision, recall, F1 score). For multiclass classification tasks, macro-averaged values were recorded.

Results Of 222 studies screened, 21 met inclusion criteria. Among these, 18 examined Parkinson's disease, 3 examined essential tremor, and 2 evaluated cerebellar ataxia. Across studies, 46 ML models were identified. The most frequently used were CNN-based architectures, including VGG16 (n=8), AlexNet (n=5), ResNet50 (n=5), and VGG19 (n=4). Employing pretraining via the ImageNet database and original hybrid classification pipelines commonly demonstrated high diagnostic capacity.

Conclusions ML analysis of Archimedes spirals represents a promising, low-cost, and non-invasive approach for movement disorder diagnosis. Despite its established use in various tremor syndromes, there remains a significant underutilization of this technique for evaluation of cerebellar ataxia, underscoring the need for dataset expansion and validation to enable broader clinical adoption. In this context, we propose the creation of our original database consisting of spirals collected from ataxia patients for use with ML analysis. Pen-and-paper spirals will be collected from seated patients under standardized protocol that encompasses handedness, collection environment, and other relevant variables.

Authors/Disclosures
Jessica Martin PRESENTER	Miss Martin has nothing to disclose.
Sanaz Attaripour, MD, FAAN (University of California, Irvine)	Dr. Attaripour has nothing to disclose.
Kotaro Tsutsumi, MD	Dr. Tsutsumi has nothing to disclose.

��ɫ��

��ɫ��