ABSTRACT
Parkinson’s disease (PD) affects the dopaminergic (dopamine-producing) neurons in a particular part of the brain called substantia nigra and causes neuron affected disorder. Single Photon Emission Computed Tomography (SPECT) for dopaminergic imaging has helped a lot in the early detection of PD. It has been found that 10% of patients diagnosed as PD show normal SPECT scans, which are called Scans without Evidence of Dopaminergic Deficit (SWEDD) and are found to be healthy in further check-ups. In this chapter we have tried to classify the SPECT scans using Convolutional Neural Networks which have been shown to perform well in complex image classification problems. We show that binary classification using CNN (PD and Healthy or SWEDD) gives competitive results compared to relevant work in this area. In this chapter, we have also attempted multi-class classification (PD, Healthy and SWEDD) and found that CNN performs well in this case, too. We used SPECT images from ninety-eight PD, ninety-two SWEDD and 102 Healthy subjects from the Parkinson's Progression Markers Initiative (PPMI) database, which are passed through a pre-processing step and then to the CNN classifier, which learns the necessary features during the training process and then classifies the image into one of the classes using those features.
