# End-to-End Parkinson Disease Diagnosis using Brain MR-Images by 3D-CNN

Abstract - In this work, we use a deep learning framework for simultaneous classification and regression of Parkinson disease diagnosis based on MR-Images and personal information (i.e. age, gender) using 3D-Convolutional Neural Networks. We intend to facilitate and increase the confidence in Parkinson disease diagnosis through our deep learning framework.

Keywords: Parkinson disease, Brain MR-Image, Deep-learning, Computer vision

Highlights

Fig.1. Parkinson disease dataset overview; HC: Healthy condition, PD: Parkinson disease

Fig.2. (a) full and (b) skull-stripped brain MR-Image - from left to right: Coronal, Axial, and Sagittal views

Fig.3. Architecture of the 3D-Convolutional Neural Network model: $L_0$: MR-Image $(80×100×108×1)$; $L_1, L_2$: Conv. $(3 3 \times 32)$; $L_4, L_5$: Conv. $(33 \times 64)$; $L_7, L_8$: Conv. $(33 \times 128)$; $L_3, L_6, L_9$: Max-pool $(2^3, 4^3, 4^3)$; $L_{10}, L_{11}$: F.C. $(512 ,128)$; $L_{12}$: Output (c) - same padding for $L_1, L_2, L_4, L_5, L_7, L_8$, and strides of two for max-pooling layers.

Fig.4. Brain’s Heat-map for Parkinson diagnosis - from left to right: (a) Coronal, (b) Axial, and (c) Sagittal views - In both Coronal and Axial views we see that Basal Ganglia and Substantia Nigra (bottom blue regions) together with Superior Parietal part on right hemisphere of the brain are found to be of critical importance in diagnosis of Parkinson, where the former one is completely corroborated by medical studies that when dopamine receptors in the striatum are not adequately stimulated those parts get either under- or over-stimulated and lead to Parkinson. However, our latter finding (i.e. Superior Parietal part) is a novel finding which asserts that not only the Basal Ganglia but also Superior Parietal part of the brain play role in Parkinson disease.