Dan Elton Email and Phone Number

Research on leveraging EHR, imaging, and genetics data for preventative healthcare and longevity.

Deployment and testing of AI systems in the radiology clinic. Research on AI for risk prediction, diagnosis, and triage with Prof. Christopher Bridge and others.

I worked as a contract Staff Scientist in the lab of Dr. Ronald Summers in the Department of Radiology and Imaging Sciences at the NIH Clinical Center. I was very fortunate to be involved in a large number of projects while at NIH. Here are some of things I did: • Developed a deep learning (CNN) based system for kidney stone detection and size quantification on CT scans which out-performed previous state-of-the-art on a challenging test set of noisy CT scans.• Supervised and mentored a post-baccalaureate fellow and three summer interns. • Trained a 3D U-Net ensemble for pancreas segmentation using an active learning approach which achieved state-of-the art performance on non-contrast CT. • Developed a patch-based 3D U-Net for segmentation of plaque in the aorta and pelvic arteries. The system produced plaque severity scores that correlated well with manual measurements (r^2 = 0.94).• Developed a fully automated deep learning based system for bone mineral density measurement in CT scans which utilizes an iterative instance algorithm to segment and label the entire spine.• Constructed a large dataset of MRI scans and annotations which is being used for machine learning endeavors in the lab. • Made numerous improvements to NIH C++ codes for automated bone mineral density measurement, fat measurement, and fracture detection. • Trained 3D U-Net models for liver region segmentation and spleen segmentation.• Developed a variational autoencoder architecture for 1-5 year survival prediction for opportunistic risk prediction using routine CT colonography scans. • Utilized CycleGAN and UNIT image translation models to generate synthetic non-contrast CT images to augment the training of deep learning models.• Assisted with GPU server installation, maintenance, and backups.

Worked for Prof. Peter W. Chung studying applications of machine learning to molecular design and discovery. Co-supervised by Prof. Mark Fuge.• Demonstrated for the first time that machine learning models can predict the properties of energetic materials (explosives & propellants) with high accuracy and low computational cost. Showed how ML can predict sensitivity to detonation, which is important for safety and is otherwise hard to predict in-silico. • Demonstrated how sensitivity analysis of machine learning models and feature ranking techniques can be used to help discover relationships between molecular structures and properties.• Wrote a review article on deep learning architectures for molecular generation and demonstrated how a generative adversarial network can be used to generate sets of potentially useful molecules.• Explained the utility of machine learning methods to program managers and chemists in DoD agencies. • Worked with postdoc Zous Boukouvalas on comparing the utility of PCA, ICA, and IVA for dimensionality reduction and data fusion prior to machine learning.• Supervised a masters student and four undergraduate students on a natural language processing project to extract chemical names, properties, and functionalities from large corpora of text extracted from pdfs and patent applications. Wrote code to calculate word2vec and GloVe embeddings and studied the clustering of chemical names in the word embedding space.

PhD adviser: Prof. Marivi Fernandez-SerraThesis title : "Understanding the Dielectric Properties of Water" I wrote and published four journal articles, gave six conference talks, and presented eight poster presentations.

Worked with Dr. Garrett Kenyon on biologically-inspired neural networks for computer vision.

Worked with Dr. Peter Persans doing experimental physics research on photothermal deflection spectroscopy.

Worked with Prof. Miriam Forman on solar wind data analysis.