CCE Theses and Dissertations

Date of Award


Document Type


Degree Name

Doctor of Psychology (PhD)


College of Computing and Engineering


Sumitra Mukherjee

Committee Member

Michael J. Laszlo

Committee Member

Frank J. Mitropoulos


Alzheimer, biomarkers, convolutional neural networks, deep neural networks, image classification


Cerebral microbleeds (CMB) are small foci of chronic blood products in brain tissues that are critical markers for cerebral amyloid angiopathy. CMB increases the risk of symptomatic intracerebral hemorrhage and ischemic stroke. CMB can also cause structural damage to brain tissues resulting in neurologic dysfunction, cognitive impairment, and dementia. Due to the paramagnetic properties of blood degradation products, CMB can be better visualized via susceptibility-weighted imaging (SWI) than magnetic resonance imaging (MRI).CMB identification and classification have been based mainly on human visual identification of SWI features via shape, size, and intensity information. However, manual interpretation can be biased. Visual screening may miss small CMB or be confused by CMB mimics. Therefore, developing automatic methods for CMB detection is critical, and recent research has been directed at finding solutions based on automated feature extraction. One of the most promising automated solutions uses a 3-dimensional convolutional neural network (3D-CNN) approach to screen and discriminate CMBs. The method uses an improved sliding window strategy, avoiding redundant computation and reducing the classification workload. However, despite its satisfactory results, the technique still suffers from limitations such as the lack of spatial information, poor handling of CMB size variation, and the existence of CMB mimics. This dissertation implements the SPP strategy into the 3D CNN two-stage model. It investigates its advantages over extant methods for CMB detection to enhance its discrimination capabilities without compromising detection speeds. The suggested model improved the results obtained by the 3D-CNN method yielding an overall 96.94% sensitivity and 95.48% precision. Another contribution is providing the 3D-CNN-SPP with the capability to detect CMBs of different sizes and shapes.