This paper reports a soft magnetoelastic sensor plus 1D-CNN and smartphone app that converts hand tremors into high-fidelity electrical signals to classify simulated Parkinsonian tremors with 98.36% accuracy.

The device offers an accessible, real‑time tremor monitoring and diagnostic tool that can aid patient stratification and remote data collection for clinical studies, but it provides minimal insight into PD pathophysiology or direct therapeutic targets.

Parkinson's disease (PD) affects over 10 million people globally. Yet, its diagnosis, neuropathology, and treatment remain challenging. PD primarily impacts the elderly, often necessitating complex medication regimens and increasing the risk of medication errors and overdoses. In this study, we report a machine-learning-assisted soft magnetoelastic ball for PD diagnosis. Utilizing the magnetoelastic effect and magnetic induction, it converts hand tremors into high-fidelity electrical signals with a 50 ms response time and a 64.6 dB signal-to-noise ratio. With 300% stretchability and Young's modulus of 214.53 kPa, it maintains a pressure sensitivity as low as 0.95 kPa. The device underwent ten simulated PD features before implementation with a one-dimensional convolutional neural network to distinguish, achieving 98.36% classification accuracy. A customized smartphone application was integrated to record and transmit data to clinicians, enabling real-time, data-driven PD diagnostics. This inclusive tool enhances accessibility for individuals with limited technological proficiency or visual, verbal, hearing, or writing impairments.