Objective: To use automated image analysis framework for tremor feature extraction from Archimedes spirals and analyse the findings based of the tremor’s type.

Background: Defining tremor’s type can be challenging and requires electrophysiological studies, which may not be available. Thus, we propose an artificial intelligence (AI)-based model for tremor analysis as a more affordable and accessible alternative.

Method: We recruited patients with tremor (Essential tremor (ET), ET plus, parkinsonian and Cerebellar tremor). We tested Archimede spiral for both hands while recording drawing time. We developed an image processing pipeline to analyze spiral drawings. A spiral flattening algorithm converted the 2D spiral into a linear form. We computed deviation metrics, total arc length, velocity, directional changes, and amplitude. A Fast Fourier Transform (FFT) filtered within 3.5–7 Hz extracted the tremor frequency.

Results: Twenty-six spirals were implemented to the AI model (8 Essential Tremor(ETS), 4 ET plus, 2 cerebellar spirals (CS), 12 parkinsonian tremor (PS)). Mean Standard deviation in ETS is 38,6 pixels(px);  17,4 in ET plus; 25 in PS and 24,6 in CS. The average of arc length was 3373 px in ETS; 2727 in CS; 2804 in PS and 1653 in ET plus. Mean velocity results were: 190,4 px per second in ETS; 238,9 in ET plus; 327 in PS and 27731 in CS. Median of amplitudes were 230,2 px in ETS; 117,36 in ET plus; 155,3 in PS and 158,41 in CS. In terms of dominant frequency, ET spiral had a mean frequency of 4,26Hz, 3,99 in ET plus; 4.43 in PS and 5,54 in CS.

Conclusion: Our AI-model helped extract features which are in accordance with tremor types. Standard deviation and arc length are higher in ET and CS, reflecting their higher amplitude found in our results. Velocity is highest in CS, highlighting the large variations and sudden accelerations and decelerations. Dominant frequencies are in line with established frequencies of each tremor. Greater amplitude and arc length in ETS compared to ET plus is probably due to the soft signs that are present among ET plus patients, reducing amplitude so making shorter arc.

processing of Archimedean spiral drawings

« Back to 2025 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/affordable-aspect-of-artificial-intelligence-applied-to-tremors-analysis-features-extraction-approach-based-on-archimedes-spiral-processing/