Parkinson’s disease is a central nervous system disorder that affects the motion of the human body that include shaking, rigidity, slow movement and difficulty in walking. As the disease advances, thinking and behavioral problems start taking place that affects a person’s psychology and ultimately leads to depression. Parkinson’s disease affects about 200,000 people on an average annually. It is an incurable disease but can be contained with treatment. Parkinson’s disease can be diagnosed on the basis of the symptoms that include muscle spasms, stiffness, movement coordination problems, dizziness, impaired speech, voice box spasms etc. Ways to diagnose it is based on the patient’s medical history as well as neurological examination. There is no lab test that will clearly identify it
and brain tests are used to rule out other diseases. This is the work on the speech pattern and gait pattern recognition on basis of which Parkinson patients and healthy patients can be classified as the motor and speech are the parts that have the most effects.
There are various tools available for investigating the time-series data of the patients obtained at clinical research centers. For the linear and nonlinear analysis of signals, using linear and nonlinear dynamic parameters, several software packages such as CDA (Chaos Data Analyzer Programs), NLyzer (Nonlinear Analysis in Real Time) TISEAN (Nonlinear Time Series Analysis), WFDB Software, MATLAB Software and Physio Toolkit Software are available. Here we have used Machine Learning techniques to classify the walk pattern and speech pattern data in order to obtain a clear classification between the type of patients using the inbuilt functions and toolboxes available in
MATLAB and Python. Using ANN, Fuzzy Logic, SVM, K-Means, AdaBoost etc. on the EMG and EEG Data of the patients we can classify them and do an early diagnose.
Using the methods, an easier and faster way of diagnosing Parkinson’s disease was found. This way an initial way of disease identification can be done based on simple EEG and EMG data. This can save a lot of time, efforts and money. A patient’s health can be analyzed by saving long-term costs. The expected results will help the research community work more on learning the various patterns that can be observed in the different data collected from sensors for different types of patients. This can help the community grow more complex algorithms to identify the difference in more efficient way. Also a web-based or mobile-application based real time data can be fetched, analyzed and prediction can be made after using the training dataset to train the models.
Whilst working on the dataset there might be several complications that may arise due to various factors. For example, multiple datasets for a given disorder often exist, collected from different sources and using slightly different features. Combining them in some effective way into a large, cohesive dataset would result in a more robust and well-trained learner.