Parkinsons-Disease-Detection

Overview

Worked on a Medical Computer Vision project involving Parkinson's Disease Detection by using Histogram of Oriented Gradients (HOG), Machine Learning and OpenCV on the images generated by the Spiral-Wave test. Detected non-uniform patterns and distortions in handwriting through the Spiral-Wave tests and classified images as Parkinson's or Healthy. Used Random Forest Classifier for Spiral images in the dataset and KNN for Wave images along with Histogram of Oriented Gradients (HOG) for quantifying the images before training. Achieved 86.66% accuracy for Spiral and 76.66% accuracy for Wave images in the dataset

Preprocessing & Training

The following preprocessing was applied to each image:

• Have trained the network on frontal handwritten images
• Resized every image to 200 × 200 pixels from the input images of random sizes
• Converted each image from RGB to GrayScale to have a single channel using cv2.cvtColor
• Thresholding the image so that it appears as white on a black background for better feature extraction using cv2.threshold
• After this , HOG was used to extract features from the images by using feature.hog function
• For Spiral : RandomForestClassifier was used for fitting & For Wave : KNeighborsClassifier was used.

Libraries Used

1.OpenCV
2.sklearn
3.skimage
4.Numpy
5.Seaborn
6.Matplotlib
7.Imutils

Results

Spiral Test Accuracy : 86.66%

Wave Test Accuracy : 76.66%

Contributors

-Rohan Limaye: https://github.com/rylp
-Rohan Naik: https://github.com/rohan-naik07