Objective: To predict tremor severity using the Archimedes spiral drawing, a commonly used item in tremor assessment scales.

Background: The Archimedes spiral is widely used in clinical tremor assessments, including the Tremor Research Group Essential Tremor Rating Assessment Scale (TETRAS). It is a quick and valuable tool for assessing action tremor. Its utility could be enhanced by using artificial intelligence (AI) and machine learning models to extract tremor features and predict severity instantly from the spiral drawing.

Method: We collected data from patients with cerebellar, parkinsonian, essential, and essential tremor plus syndromes. Tremor severity was assessed using TATRAS, and spiral drawings were analyzed for both hands while recording drawing time. The AI model transformed the two-dimensional tracing of the Archemides spiral into a single horizontal line,a technique called “spiral flattering”. This simplifies tremor analysis and easily extracts tremor features: standard deviation, arc length, dominant frequency, mean velocity, and mean amplitude. We used the Support Vector Regression (SVR) as a non-linear regression AI model for our prediction task. To ensure the reproducibility of our results, we train our model using 4-fold cross-validation techniquefor training  and the left-out fold is used for testing. At the end of the cross-validation, we will have the list of the predicted scores for all patients. Then, we calculate the absolute distance between ground truth and predicted clinical scores and average it across all patients, resulting in the mean absolute error (MAE).

Results: Thirty-two spirals were used to train our SVR model. The MAE values for different tasks were: 0.54 for the pouring task, 0.79 for carrying food, 0.81 for using keys, 0.81 for writing, 2.3 for total upper limb tremor score, 5.6 for the daily activities subscale, 6.2 for the total performance subscale, 10.2 for the total TETRAS score.

Conclusion: Using AI and machine learning, particularly SVR, to predict tremor severity from an instant photo of the Archimedes spiral appears to be a promising, affordable, and time-efficient tool for clinicians. This approach could allow rapid assessment of specific tasks such as writing, pouring, or holding keys, enhancing clinical decision-making.

Figure 1: Machine learning model pipeline

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/archimedes-spiral-based-non-linear-regression-machine-learning-model-for-predicting-tremors-severity/