Subclinical tremor differentiation using long short-term memory networks

Area of Science:

• Neurology
• Biomedical Engineering
• Artificial Intelligence

Background:

• Subclinical tremors in Parkinson's disease (PD) and essential tremor (ET) present diagnostic challenges, even with apparent tremors.
• Up to 30% of PD cases have subclinical tremors, a poorly understood phenomenon.
• Accurate differentiation is crucial for timely diagnosis and treatment planning.

Purpose of the Study:

• To develop and evaluate a deep learning model for differentiating subclinical tremors.
• To distinguish between tremors associated with PD, ET, and normal physiological tremors.
• To leverage artificial intelligence (AI) for improved diagnostic accuracy in movement disorders.

Main Methods:

• Utilized inertial sensor data from hands and arms of 51 PD, 15 ET, and 58 normal subjects.
• Developed a deep learning model using a long short-term memory (LSTM) network.
• Trained the LSTM network on short-time Fourier transformed subclinical tremor data.

Main Results:

• The LSTM model achieved 95% accuracy differentiating PD and ET tremors.
• The model achieved 93% accuracy differentiating PD, ET, and physiological tremors.
• The proposed method demonstrated 30-50% higher accuracy for low-amplitude tremors compared to existing methods.

Conclusions:

• The AI-driven LSTM model shows significant potential for differentiating subclinical tremors.
• This approach can enhance diagnostic accuracy for Parkinson's disease and essential tremor.
• Future work includes model interpretability and validation on larger, diverse datasets.