Advisors:
Barbara Boyan, PhD, Dean of the School of Engineering, Virginia Commonwealth University
Brani Vidakovic, PhD, Department of Biomedical Engineering, Georgia Institute of Technology

Thesis Committee:
Zvi Schwartz, DMD, PhD, School of Engineering, Virginia Commonwealth University
Douglas Robertson, MD, PhD, Department of Radiology, Emory University
Anthony Yezzi, PhD, School of Electrical and Computer Engineering, Georgia Institute of Technology

The overall objective is to develop an automated method to analyze human craniosynostosis using standard CT scans and quantify disease presence and progression for clinical assessment using predictive modeling. We hypothesize that a new automated method to evaluate CT scans will allow us to quantify cranial development to identify and distinguish different types of craniosynostosis as a clinical diagnostic aid, from onset to full presentation of symptoms.

Specific Aim 1: To characterize intracranial volume asymmetries and cranial suture measurements for synostosis differentiation and quantification using low-dose cranial CT scans. The objectives of this specific aim were to (i) analyze total intracranial volume and cranial asymmetry using graphical user interfaces developed in MATLAB, (ii) characterize suture specific responses to different synostosis types by measuring bone distance, volume, and percentage open, and (iii) determine age variations and fontanel effects on suture measurements. The working hypothesis is that intracranial volume ratios and suture measurements can be used to quantify the type of synostosis and severity.

Specific Aim 2: To develop and validate a predictive model for determination of probability and severity of each synostosis type using logistic regression. The objectives of this specific aim were to (i) construct the regression models using feature selection to obtain the most relevant feature set for prediction of the type of synostosis and (ii) validate the regression models using pseudo R-square metrics and cross-validation to assess performance and scalability. The working hypothesis is that logistic regression models can be developed to identify the onset and severity of craniosynostosis for future disease quantification.