Enhancing PD Diagnosis through Handwriting Analysis: A CNN-BiLSTM Model with Dual Feature Extraction and GOOSE Optimization

Motor symptoms like stiffness, tremor, bradykinesia, and akinesia are typical of Parkinson's disease (PD). To aid in evaluation of PD, it is useful to analyze patients' fine motor control, especially their handwriting. There has been extensive research on the use of online handwriting acquisition techniques to detect PD by analyzing a variety of dynamic handwriting parameters. It is well-known that all current methods have their advantages and disadvantages, and that there has been much research into hand motion detection of single-modality data for the study of different contact besides noncontact sensors. This research presents a new deep learning model that integrates CNN, LSTM, and Bi-directional LSTM as a hybrid model for PD diagnosis through the use of handwriting images. With low resolution comes the intrinsic issue of balancing spatial and angular resolution. To solve this, this research offers a Feature Extraction (FE) approach. In both the initial and deep feature extraction phases, we present a novel method for picture reconstruction using a convolutional neural network (CNN) that integrates spatial and epipolar data. In order to efficiently assess both horizontal and vertical epipolar data, our network makes use of angle information during upsampling and uses dual feature extraction. The GOOSE algorithm then selects the most pertinent features, which increases the accuracy of the classification. Additionally, when compared to comparable research in PD detection, the experimental results show that the suggested approach obtains greater identification rates.