DEEP LEARNING BASED PARKINSON’S DISEASE DIAGNOSIS THROUGH RECURRENT NEURAL NETWORKS WITH INCEPTION NET MODELS

Authors: Mrs. Eurekha, Mrs. Samundeeswari

Abstract: Parkinson’s Disease (PD) is a progressive neurodegenerative disorder that affects millions globally, characterized by motor and non-motor impairments. Early diagnosis is essential to improve patient outcomes but remains a challenge due to subjective clinical assessments and delayed symptom recognition. This study proposes a novel, automated diagnostic framework leveraging deep learning techniques, particularly the integration of InceptionNet and Long Short-Term Memory (LSTM) networks. The system processes SPECT imaging data to extract spatial features using InceptionNet and captures temporal patterns via LSTM networks. Bilinear pooling is employed to fuse spatial and sequential features, enhancing classification performance. Evaluated on public handwriting datasets such as PaHaW and DraWritePD, the model achieved a diagnostic accuracy of 91.7%, with a high F1-score (90.9%) and AUC (0.94), outperforming traditional methods. This hybrid architecture offers a non-invasive, reliable, and scalable solution for early PD detection, demonstrating strong potential for integration into clinical workflows. Future improvements may include real-time diagnostics, explainable AI, and cross-platform deployment to support large-scale medical applications.

DOI: