Linear Gaussian Kernel Cycle Generative Adversarial Networks based Transformer Learning for Early-Stage Stroke Prediction with Parkinson and Skin Disease

Research overview

The text below is the source article’s original abstract under a verified reuse license. cc by

Abstract Stroke is the main cause of mortality and long-term disability in the whole world, and therefore, early prediction of the condition is important to enable timely medical care. Recent reports suggest that neuro-related diseases like Parkinson diseases and skin related diseases like skin wrinkles may offer reward valuable information on measuring stroke risk. Nevertheless, the current deep learning models are usually affected by difficulty in attaining high prediction accuracy and low computational time. In order to overcome these weaknesses, this paper introduces a new model of Linear Gaussian Kernel Cycle Generative Adversarial Networks-based Transformer Learning (LGKCGAN-TL) to predict strokes at an early stage. The model proposed incorporates several processing frames such as image capture, preprocessing, data augmentation, features extraction, and classification. Firstly, the pictures of Parkinson drawings and Dermnet skin disease datasets are gathered. To improve the quality of an image, the index box linear denoising technique of Camargo is used to process the image to reduce all the noise artifact and maximize the peak signal to noise ratio. This is followed by the use of Gaussian Kernel Cycle Generative Adversarial Networks (GANs) to increase the dataset and create new training samples and enhance the robustness of the model. In the case of feature learning and classification, a Rosenthal Correlative Transformer Learning model is adopted to extract geometric, texture, color, and shape features of medical images and achieve correct stroke risk classification. There is a vast experimental assessment with such metrics as PSNR, accuracy, precision, recall, F1-score, specificity, ROC analysis, and stroke prediction time. Experimental outcomes prove that the suggested LGKCGAN-TL model reaches a higher performance of 98-99 on average prediction accuracy, 98.5 on precision, 97.8 on recall, F1-score of 98 and specificity of 93, halting the stroke prediction time to about 55 ms, which is better than other models based on GA-BiLSTM and DCNN.

Related topics