The current focus of my work is extending a recent project on the automated segmentation of the vascular network from digital images of the placenta. Morphology differences in the placental chorionic surface vascular network have been associated with higher risk of developmental disorders such as autism spectrum disorder (ASD). "[P]lacentas of high-risk ASD pregnancies generally had fewer branch points, thicker and less tortuous arteries, better extension to the surface boundary, and smaller branch angles than their population-based counterparts." (J.-M. Chang et al.) Thus, a composite of such features could potentially be used as a non-invasive, early diagnosis tool. Furthermore, it is important to understand the mechanisms which drive these changes - are they causal in the development of ASD or a result of another, underlying factor?
I am using deep learning methods (such as the beautiful work of Isola et al.) to complement mathematical methods of image processing for vessel extraction. Currently, the tracing is done manually by a trained researcher, a process that is time-consuming and costly. The automated detection of vessels would permit large-scale studies for research and clinical applications.
Previous ProjectsA list of my publications can be found on my Google Scholar profile.
ISOpureR. During my first post-doctoral project, I worked on identifying the fraction and molecular profiles of normal and cancer cells within a tumor. I translated and optimized a deconvolution algorithm. The ISOpureR code is on CRAN.
Epilepsy Seizure Prediction (Kaggle Competition). The challenge was to distiguish between ten minute iEEG data clips recorded in the hour before a seisure from normal activity. I used a simple approach of an ensemble of different models.
Alzheimer's Disease Big Data DREAM Challenge #1. We worked on the first two parts of the challenge, using clinical variables to predict cognitive decline and to predict which individuals have normal cognitive function despite indicators of amyloid deposition.