Assessing parkinsonism & cerebellar dysfunction with spiral & line drawings

Area of Science:

• Neurology
• Biomedical Engineering
• Machine Learning

Background:

• Accurate diagnosis of neurological diseases remains a significant challenge.
• Speech, movement, and drawing are explored as diagnostic modalities.
• Distinguishing between Parkinson’s and cerebellar symptoms requires reliable methods.

Purpose of the Study:

• To compare the efficacy of spiral and line drawings in recognizing Parkinson’s and cerebellar symptoms.
• To evaluate the impact of pin pressure data on classification accuracy.
• To assess the benefit of combining spiral and line drawings for improved diagnosis.

Main Methods:

• Generating image data from raw drawings, with and without pressure information.
• Utilizing pre-trained and custom deep learning models for feature extraction and classification.
• Applying the Mann-Whitney U test to determine statistical significance (p < 0.05).

Main Results:

• Spiral drawings demonstrated significantly higher recognition performance compared to line drawings (p=0.001).
• Combining both drawing types improved classification accuracy when pressure data was included (p=0.017).
• Classification performance showed no significant degradation when pressure data was omitted for a single drawing task (p=0.507).

Conclusions:

• Spiral drawings are recommended as a primary method for neurological symptom recognition.
• Integrating multiple drawing types can enhance diagnostic confidence, particularly with pressure data.
• Pressure data is not critical for maintaining classification performance when using a single drawing task.