Elasticnet-Based Vision Transformers for early detection of Parkinson's disease

Abstract

Parkinson's disease is a progressive neurodegenerative disease that is very difficult to diagnose early. The use of Traditional Convolutional Neural Network-based approaches is becoming increasingly common in health problems such as Parkinson's disease, where early diagnosis is vital. In this study, Vision Transformer and Convolutional Neural Network models were used in a comparative and innovative approach for the early diagnosis of Parkinson's disease with 427 faster and 99.9% accuracy rate. Images obtained from the hand drawings of Parkinson's patients were trained with the current convolutional neural network models EfficientNet and SqueezeNet variations. In addition, feature extraction is performed with Vision Transformer models, and classification is performed through machine learning techniques. In the proposed model, the features extracted from the Vision Transformer models are filtered using the ElasticNet feature selection algorithm, and model training is performed using machine learning classifiers. The results obtained show that Vision Transformer models can diagnose Parkinson's disease over handwriting data with higher speed, accuracy, and precision compared to traditional deep learning methods. The most successful result in model training with the proposed method was obtained with an accuracy rate of 99.9% from the models trained with the features taken from the Vit_B_16 architecture from Vision Transformer models. When the model training is analyzed, it is concluded that the proposed method has higher speed, accuracy, and precision compared to traditional deep learning models.