A COMPARATIVE STUDY USING ENSEMBLE NETWORKS FORMED FROM STATE-OF-THE-ART FINE-TUNED DEEP CONVOLUTION NEURAL NETWORKS FOR THE ACCURATE DETECTION AND CLASSIFICATION OF PROGRESSIVE SUPRA-NUCLEAR PALSY

For the diagnosis and classification of Progressive Supranuclear Palsy (PSP), a subtype of Parkinson’s disease (PD), accurate and rapid computer-assisted magnetic resonance imaging (MRI) assessments are necessary. The symptoms of this neurodegenerative disease appear fast and worsen. PSP clinical symptoms overlap, making diagnosis difficult. Thus, computer-assisted MR image evaluations are needed to diagnose and classify PSP. The main drawback of PSP is the absence of comprehensive research, unlike other neurological diseases. This work presents an enhanced machine-learning strategy for PSP progression prediction. From several databases, 450 high-resolution T1-weighted MRIs were retrieved for this investigation. A total of 150 photographs of people with PSP, 150 people with PD, and 150 healthy people without neurological disorders served as control subjects in this study. A number of pretrained TL models, an optimized CNN model, a hybrid LSTM/CNN model, and optimal Ensemble models trained with fine-tuned TL models and an optimized CNN model are all going to be tested in this prospective research. Using these models, we can differentiate between PSP patients, PD patients, and control groups. A self-governing hyperparameter optimization strategy is used to enhance the effectiveness of the customized CNN model. The custom model CNN research work ensemble model, which combined the VGG19 custom model and the Inception Resnet v2 custom model, outperformed both the individual models and other ensembles. In our study, authors evaluated the model’s categorization accuracy to other models and scholarly literature. PSP patients were classified 96% differently from control patients and 99% differently from PD patients by the model. Recipient operating characteristic curves support the ensemble learning process. This new work uses complex ensemble models to detect and classify PSP from PD and control patients in a generalizable, sensitive, and variance-reducing fashion. This cutting-edge research can examine other ailments.