A convolutional-recurrent neural network approach to resting-state EEG classification in Parkinson’s disease

Abstract

Background: Parkinsons disease (PD) is expected to become more common, particularly with an aging population. Diagnosis and monitoring of the disease typically rely on the laborious examination of physical symptoms by medical experts, which is necessarily limited and may not detect the prodromal stages of the disease. New Method: We propose a lightweight (20K parameters) deep learning model, to discriminate between resting-state EEG recorded from people with PD and healthy controls. The proposed CRNN model consists of convolutional neural networks (CNN) and a recurrent neural network (RNN) with gated recurrent units (GRUs). The 1D CNN layers are designed to extract spatiotemporal features across EEG channels, which are subsequently supplied to the GRUs to discover temporal features pertinent to the classification. Results: The CRNN model achieved 99.2% accuracy, 98.9% precision, and 99.4% recall in classifying PD from healthy controls (HC). Interrogating the model, we further demonstrate that the model is sensitive to dopaminergic medication effects and predominantly uses phase information of the EEG signals. Comparison with Existing Methods: The CRNN model achieves superior performance compared to baseline machine learning methods and other recently proposed deep learning models. Conclusion: The approach proposed in this study adequately extracts the spatial and temporal features in multi-channel EEG signals that enable the accurate differentiation between PD and HC. It has excellent potential for use as an oscillatory biomarker for assisting in the diagnosis and monitoring of people with PD. Future studies to further improve and validate the model performance in clinical practice are warranted.

Detection of Parkinson’s disease from EEG signals using discrete wavelet transform, different entropy measures, and machine learning techniques

“…As we mentioned earlier, the aim of our study is to provide an efficient and, at the same time, less complex model than those found in previous studies. There is no doubt that deep learning-based models [26][27][28][29][30] offer promising results, but at the cost of simplicity. For example, the number of trainable parameters reached 100 K in 30 , 20 K www.nature.com/scientificreports/ in 28 , and 6602 in 26 .…”

“…There is no doubt that deep learning-based models [26][27][28][29][30] offer promising results, but at the cost of simplicity. For example, the number of trainable parameters reached 100 K in 30 , 20 K www.nature.com/scientificreports/ in 28 , and 6602 in 26 . The model in study 27 (CNN + LSTM) used the lowest number of parameters, which was 380.…”

“…In terms of resource utilization, deep learning techniques need high-performance memory and processors. For example, the model developed in 28 (CNN + RNN) has been implemented with a GPU machine, as the authors reported. The authors of 29 stated that their developed model necessitates a large amount of computer memory.…”

“…The CNN-based model in 26 was executed on a computer with two Intel Xeon 2.40 GHz processors and a 24 GB random access memory. Regarding the execution time, the authors of 28 reported that their model required 15 min for training, but once it has been trained, it takes less than a second to test a new segment. The authors of 29 also reported that their model (2D-CNN) is computationally demanding, which results in long training times.…”

Mental state diagnosis using Hybrid CNN with EEG signal from wearable device

“…The proposed CRNN algorithms extract spatiotemporal features from EEG channels to GRUs to detect the diseases. The cross validation and feature extraction layers will increase the complexity of classification [11]. There are many wavelet based neural networks to predict the state of mind of a person.…”