Date on Master's Thesis/Doctoral Dissertation


Document Type

Doctoral Dissertation

Degree Name

Ph. D.


Electrical and Computer Engineering

Degree Program

Electrical Engineering, PhD

Committee Chair

El-Baz, Ayman

Committee Co-Chair (if applicable)

Inanc, Tamer

Committee Member

Inanc, Tamer

Committee Member

Nasaroui, Olfa

Committee Member

Faul, Andre

Author's Keywords

autism; MRI; CAD system


The term “autism spectrum disorder” (ASD) refers to a collection of neuro-developmental disorders that affect linguistic, behavioral, and social skills. Autism has many symptoms, most prominently, social impairment and repetitive behaviors. It is crucial to diagnose autism at an early stage for better assessment and investigation of this complex syndrome. There have been a lot of efforts to diagnose ASD using different techniques, such as imaging modalities, genetic techniques, and behavior reports. Imaging modalities have been extensively exploited for ASD diagnosis, and one of the most successful ones is Magnetic resonance imaging(MRI),where it has shown promise for the early diagnosis of the ASD related abnormalities in particular. Magnetic resonance imaging (MRI) modalities have emerged as powerful means that facilitate non-invasive clinical diagnostics of various diseases and abnormalities since their inception in the 1980s. After the advent in the nineteen eighties, MRI soon became one of the most promising non- invasive modalities for visualization and diagnostics of ASD-related abnormalities. Along with its main advantage of no exposure to radiation, high contrast, and spatial resolution, the recent advances to MRI modalities have notably increased diagnostic certainty. Multiple MRI modalities, such as different types of structural MRI (sMRI) that examines anatomical changes, and functional MRI (fMRI) that examines brain activity by monitoring blood flow changes,have been employed to investigate facets of ASD in order to better understand this complex syndrome. This work aims at developing a new computer-aided diagnostic (CAD) system for autism diagnosis using different imaging modalities. It mainly relies on making use of structural magnetic resonance images for extracting notable shape features from parts of the brainthat proved to correlate with ASD from previous neuropathological studies. Shape features from both the cerebral cortex (Cx) and cerebral white matter(CWM)are extracted. Fusion of features from these two structures is conducted based on the recent findings suggesting that Cx changes in autism are related to CWM abnormalities. Also, when fusing features from more than one structure, this would increase the robustness of the CAD system. Moreover, fMRI experiments are done and analyzed to find areas of activation in the brains of autistic and typically developing individuals that are related to a specific task. All sMRI findings are fused with those of fMRI to better understand ASD in terms of both anatomy and functionality,and thus better classify the two groups. This is one aspect of the novelty of this CAD system, where sMRI and fMRI studies are both applied on subjects from different ages to diagnose ASD. In order to build such a CAD system, three main blocks are required. First, 3D brain segmentation is applied using a novel hybrid model that combines shape, intensity, and spatial information. Second, shape features from both Cx and CWM are extracted and anf MRI reward experiment is conducted from which areas of activation that are related to the task of this experiment are identified. Those features were extracted from local areas of the brain to provide an accurate analysis of ASD and correlate it with certain anatomical areas. Third and last, fusion of all the extracted features is done using a deep-fusion classification network to perform classification and obtain the diagnosis report. Fusing features from all modalities achieved a classification accuracy of 94.7%, which emphasizes the significance of combining structures/modalities for ASD diagnosis. To conclude, this work could pave the pathway for better understanding of the autism spectrum by finding local areas that correlate to the disease. The idea of personalized medicine is emphasized in this work, where the proposed CAD system holds the promise to resolve autism endophenotypes and help clinicians deliver personalized treatment to individuals affected with this complex syndrome.