To develop and validate an automated, quantitative algorithm and compliant digital pipeline for extracting and analyzing tremor features from Archimedes spirals to assist in clinical practice and studies in adults with tremor.
Conventional clinician-based scoring of spirals (e.g. TETRAS Performance Subscale) demonstrates poor inter-rater reliability and low sensitivity to treatment effects, with agreement across raters as low as 4%. Objective digital analysis offers a scalable, reproducible alternative. The Essential3 Spiral Analysis Initiative aimed to standardize spiral acquisition and digitization, and develop an AI-enabled framework to quantify tremor severity and predict functional improvement.
Feature extraction and algorithm training were performed using spiral drawings from healthy volunteers in the Establishment of an Archimedes Spiral Normative Dataset for Clinical Research Study to provide a normative reference for motor variability.
Participants in two randomized, double-blind Phase 3 studies assessing the impact of ulixacaltamide in essential tremor (parallel-design and randomized-withdrawal; Essential3) drew large and small Archimedes spirals with their dominant hand at baseline, Day 56, and Day 84.
Scanned spiral drawings were processed through a pipeline performing automated template removal, skeletonization, and extraction of 12 quantitative features reflecting tremor amplitude, irregularity, and complexity.
Machine-learning models were trained on combined datasets to predict treatment assignment and clinical benefit (mADL11 change).
Automated processing achieved near-complete capture of spiral drawings, generating an immutable, quality-controlled patient dataset. Quantitative metrics correlated strongly with mADL11 change and accurately predicted responder status. Algorithmic scoring demonstrated complete reproducibility, with superior sensitivity to treatment-related changes compared to clinician-based scoring.
The Essential3 Spiral Analysis Initiative established a fully automated, validated, digital pipeline transforming hand-drawn spirals into quantitative, reproducible digital features. Integration of normative data and advanced machine-learning models establishes a scalable framework for regulatory-grade digital biomarkers of motor function in patients with tremor, suitable for use in large-scale clinical trials.