Deep Learning-Based Artificial Intelligence Algorithm to Classify Tremors from Hand-Drawn Spirals.

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Objective biomarkers for diagnosing and classifying tremor syndromes are currently lacking.
• Tremor classification relies heavily on subjective clinical assessments.
• Developing objective diagnostic tools is crucial for effective patient management.

Purpose of the Study:

• To develop and validate a deep learning (DL) algorithm for classifying tremors using hand-drawn spirals.
• To assess the algorithm's performance against expert raters.
• To provide an objective, feature-independent method for tremor classification.

Main Methods:

• Recruited participants with various tremor syndromes (dystonic tremor, essential tremor, Parkinson's disease, cerebellar ataxia) and healthy volunteers.
• Utilized hand-drawn spirals to train a DL algorithm (InceptionResNetV2, Keras sequential model) via transfer learning.
• Externally validated the model on independent cohorts, comparing its accuracy and F1 scores to those of expert clinicians.

Main Results:

• The DL classifier achieved an initial overall accuracy of 81%, with an adjusted accuracy of 70% after reanalysis.
• External validation on 1535 spiral drawings yielded an accuracy of 61% (adjusted 59%).
• The DL algorithm significantly outperformed human raters, who achieved 46% accuracy.

Conclusions:

• Supervised DL algorithms can effectively detect and classify tremor syndromes from simple hand-drawn spirals.
• This approach offers unbiased, feature-independent classification, surpassing human rater performance.
• DL-based analysis of spiral drawings presents a promising objective tool for tremor diagnosis.