Abstract
The application of a radioactive tracer and following brain Single Positron Emission Computed Tomography (SPECT) is a standard technique used in neurodegenerative disease investigation. Alzheimer's disease is the most common form of neurodegenerative disease.
In this paper, a novel 3D linear classifier is developed to classify Alzheimer's disease. The classification problem is formulated as the variational task with periodic boundary conditions, which is easy to discretize and solve using Fast Fourier Transform, and therefore, the resulting learning algorithm is very fast.
Thanks to linearity of the classifier, weights obtained by 3D classifier learning are easy to visualize and bring understanding the most important features. The proposed classifier exhibits accuracy, sensitivity, and specificity of at least 90%.