arathyjeevan/parkinson-s-diagnosis

proposed early detection method for parkinson's disease using deep learning on MRI dataset

parkinson-s-diagnosis

Parkinson’s disease is a neurological disorder with more than 6 million people worldwide suffering from it.It is commonly diagnosed using clinical assessments and progression scale which usually depends on the medical practitioner’s expertise ,and accuracy varies greatly between various examiners which also takes a long time to accurately diagnose.This paper proposes to develop a computer aided diagnostic method to diagnose PD patients using MRI images of the brain ,thus reducing cross examiner variability and the time required to accurately differentiate between PD and Control subjects.

Image Enhancement:

The images present in the dataset have varying brightness,colour and noise ,to remove these unwanted elements from our training and testing images ,we apply image filtering operations and histogram equalisation for contrast enhancement ,for better identifiable features. In our study the image enhancement pipeline consists of converting the RGB image to YUV colour space, for accurate colour and features ,then the Luminance channel is filtered using gaussian blur to reduce the noise and pixelation,then contrast limited adaptive histogram equalisation is applied on the Y channel to improve the local image contrast while keeping the noise low. finally the the 3 channels Y,U and V are merged and converted to RGB colorspace for further processing

note: to use the image enhancement technique please go to either parkisinsons_first_stage.ipynb or parkinsons_final_stage.ipynb and in the calling of the dataset functio change th eparameter enhance to true

First stage :

region of interest extraction

The images that the MRI dataset provides contain multiple slides of the brain and different studies might use different thickness of the slides . the most influencing region of the brain in the detection of Parkinson’s is the SN region , we localize the sn region to classify between pd and no pd ,to improve the accuracy of the localization we separate the images containing the SN region ,In this study we use a custom CNN for automatically differentiating SN in image and no sn in image

Final stage stage :

classification

final stage of the process is the classification of the Substantia nigra region as belonging to a Parkinson’s patient or a control. A modified version of the Alex Net with activation at the last fully connected layer having a sigmoid activation for the 2 classes and 6 convolutional layers ,to improve the feature vectors as compared to manually selected vectors, which are prone to changes in orientation and intensity. An image of the Substantia Nigra region in the MRI of size 512x512 is taken as an input by the convolutional neural network which can extract the features from the image autonomously for the classification of the image into the classes as PD or control.