EARLY DETECTION OF PARKINSON’S DISEASE USING MACHINE LEARNING

Millions of people worldwide have been affected by Parkinson’s disease (PD), which is characterized by a loss of mobility and, as a result, the inability to work and move. Early detection of Parkinson’s disease can help to cure it in time. A test that involves drawing a spiral on a sheet of paper could be used to detect Parkinson’s disease in its early stages. Speed of writing and pen pressure while sketching are lower among Parkinson’s patients, particularly those with a severe form of the disease. The goal of this project was to develop an image classification algorithm to detect Parkinson’s disease using images of spirals/waves/handwriting obtained during clinical exams. Although pen pressure plays an important role in determining the presence and the extent of the disease, this project aims to find out whether images alone can be useful in detecting Parkinson’s disease.

Dataset

For this project, a publicly available dataset from Brain Disease Analysis Laboratory has been used. The subjects were asked to draw a spiral on a piece of paper placed on a digitizing tablet. Digitized signals were obtained using the tablet when pen pressure was exerted on its surface. The tablet captured the x-coordinate; y-coordinate; timestamp; pressure; tilt; and elevation. Using the x and y coordinates, images were generated as shown below:

Only 40 samples were available in this dataset which is a very less number of samples to train any deep learning or machine learning model. To tackle this problem, data augmentation was used. It is a technique that can be used to artificially expand the size of a training dataset by creating modified versions of images in the dataset. Image data augmentation is a technique that can be used to artificially expand the size of a training dataset by creating modified versions of images in the dataset. Using this, the dataset was increased to 900 images.

Methods

Two Deep learning and machine learning algorithms were used to classify the images as “Parkinson” and “Healthy”.

1. Convolutional Neural Networks (CNN):

A CNN is a deep learning algorithm that can take an image as input, assign learnable weights and biases to multiple features of the image, and distinguish them from another. For this task, a number of pre-trained models like VGGs, ResNets, Inception, etc. were used using transfer learning however, the accuracy obtained was insignificant (>40%).

2. Random Forest Classifier

It is one of the most commonly used supervised learning algorithms which is used in classification problems. Random Forest algorithm consists of several “trees” which predict the class of the input. The prediction made by each individual tree is taken into consideration and the class with the most votes is the algorithm’s final prediction. To classify images using such algorithms, the images need to be quantified first. In this project, Histogram of Oriented Gradients (HOG) has been used for image quantification. HOG quantifies changes in local gradient in the input image. It has been used to quantify how the directions of the spirals change. In other words, it will be able to identify the degree of shakiness in the images obtained from people with PD as compared to those obtained from healthy people.

Results

The approach where random forest classifier was used showed an accuracy of 96% and the one with convolutional neural networks showed an accuracy of 66%. Thus, the random forest classifier showed more promising results which is counter intuitive as CNNs tradiotionally perform better on image datasets.