Objective: This study sought to generate a standard model for essential tremor (ET) evaluation using a convolutional neural network (CNN).

Background: The Archimedes spiral is a clinical tool that aids in the diagnosis and monitoring of ET. However, spiral ratings may vary based on experience and training of the rating physician.

Method: One senior movement disorders neurologist (Neurologist 1) rated 1,653 spiral images from 46 ET patients and 75 controls using the Bain and Findley Spirography Rating Scale. Neurologist 1’s scores were used as the reference standard in the model. Spiral images from each group were partitioned into training, validation, and testing datasets, and evaluated using 3-fold cross-validation. A subset of the testing data from the first fold was re-evaluated by Neurologist 1, evaluated by a second senior movement disorders neurologist (Neurologist 2), and evaluated by our model. From these scores, Cohen’s Weighted Kappa 95% confidence intervals, F1-scores, and mean squared error (MSE) were calculated to determine if our model performs as well as two senior movement disorders neurologists.

Results: With overlapping Cohen’s Weighted Kappa 95% confidence intervals for all comparisons, no significant differences in performance were observed between our model and the two senior movement disorders neurologists. MSE and F1 scores demonstrate better agreement between our model and each neurologist individually than either neurologist had with each other.

Conclusion: We validate a CNN model that evaluates spirals with the same proficiency as two senior movement disorders neurologists. The model also demonstrates better agreement with each rater separately than each rater has with each other.

« Back to 2023 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/electronic-archimedes-spiral-neural-network-elann/