An Enhanced Detection and Segmentation of Parkinson’s Disease using Novel Deep Learning based Approaches.

Abstract

Parkinson's disease (PD) is a progressive neurodegenerative condition histologically defined by the degeneration of dopaminergic neurons in the substantia nigra pars compacta (SNpc) and the development of Lewy bodies in many brain regions. The SNpc is a compact midbrain region essential for motor coordination and movement regulation, generating dopamine, a neurotransmitter crucial for the start, speed, and fluidity of voluntary movement sequences. The etiology of the majority of Parkinson's disease cases, referred to as 'sporadic' or 'idiopathic' PD, remains elusive, however it encompasses intricate interplay between hereditary and environmental influences. Parkinson's disease is the second most prevalent neurodegenerative ailment behind Alzheimer's disease, impacting 1% of those over 60 years old and around 5% by age 85. The incidence is increasing owing to aging demographics.       
The Parkinson Disease Foundation estimates that around 10 million individuals globally are affected with Parkinson's disease, including one million in the USA, 1.2 million in Europe, and a predicted two million in China by 2030. One in 500 persons in the UK is afflicted, and this figure is anticipated to increase thrice over the next 50 years. No established disease-modifying treatment presently exists. The diagnosis of Parkinson's Disease requires the presence of bradykinesia, with either muscular stiffness, tremor, or postural instability. Approximately 20% of patients do not have a tremor. The signs of Parkinson's disease extend beyond motor deficits. Timely identification of Parkinson's Disease is crucial for delivering suitable therapy and prognosis information to patients. Nonetheless, a precise early diagnosis may be difficult due to the overlap of movement symptoms with other illnesses. Physicians diagnose Parkinson's disease by clinical assessment, mostly using data obtained from patient history and examination. Occasionally, brain imaging may be solicited to assist in corroborating the clinical diagnosis; nevertheless, there are yet no diagnostics that are entirely sensitive or specific for Parkinson’s disease. The misdiagnosis rate of Parkinson's Disease is roughly 10–25%, with an average duration of 2.9 years needed to get 90% accuracy. Autopsy remains the definitive standard for illness confirmation.              
This study developed an advanced convolutional neural network model to predict Parkinson's disease using both picture and audio data. Typically, conventional machine learning algorithms like SVM and Random Forest do not filter data repeatedly, resulting in lower prediction accuracy. Consequently, this study employs Convolutional Neural Networks (CNN), which filter data many times via neuron values, thus enhancing prediction accuracy. This study utilizes WAVE and SINE pictures of those with normal conditions and those with Parkinson's disease for imaging data, while UCI Parkinson's recorded voices serve as speech samples.