Prediction of Preoperative Scale Score of Dystonia Based on Few-Shot Learning

Abstract

As a neurological disease, dystonia mainly has symptoms including muscle stiffness, dyskinesia, tremor, muscle spasm, etc. Dystonia score plays an important role in targeted auxiliary diagnosis, treatment plan design, and follow-up evaluation of patients. In this paper, the feature information of brain lateralization is extracted from electroencephalography (EEG) signals by clustering method, while information on time domain, frequency domain, and time sequence are extracted from EEG signals and electromyography (EMG) signals. Various deep-learning models are used to predict dystonia scores. Experiments show that this method can effectively predict dystonia based on the quantitative indicators extracted from few-shot neural signals. The methodology in this paper can help doctors judge the disease more accurately, make personalized treatment plans, and assist in monitoring the treatment effect.