Adrian Vasile Dalca

PhD. Candidate, Medical Vision Group
Computer Science and Artificial Intelligence Lab
Electrical Engineering and Computer Science
Massachusetts Institute of Technology

32 Vassar St, 32-D474, Cambridge, MA, 02139. 
adalca at mit dot edu


The following lists most of the projects, research or otherwise, that I've worked on. For a more consise list of my research you can see my list of publications. I am fortunate to work with exceptional collaborators Ramesh Sridharan, Mert Sabuncu, Kayhan Batmanghelich, Ehud Schmidt, Natalia Rost, Manolis Kellis and Jonathan Rosand. I've also been very lucky to work in bioinformatics with professor Michael Brudno and in geophysics with professor Jerry Mitrovica at the University of Toronto.

Main Projects

Joint Modeling of Imaging and Genetics

One of my main interests is on joint modeling of medical imaging and genetic variants. We aim to identify clusters of genetic markers working together to affect brain structure and function, and further explain variances in behaviour traits or clinical disease.

A Computational Model for Stroke

Stroke is one of the top causes of death and debilitating injury. We are developing models for extracting important phenotypes from medical stroke imaging to aid in prediction, risk assesment and genetic exploration. Stroke is also a main applications of our imaging genetics models mentioned above.

tipiX - Rapid Visualization of Large Image Collections

tipiX is a new approach for fast and effective visualization of large image collections. This applies to both natural images as well as medical volumes in population studies. The key insight is to collapse inherently high-dimensional imaging data onto an interactive two-dimensional canvas native to a computer screen in a way that enables intuitive browsing of the image data. Several examples will get you started in various domains. The code is available on github.

  • A.V. Dalca, R. Sridharan, N.S. Rost, P. Golland. tipiX: Rapid Visualization of Large Image Collections In MICCAI-IMIC Interactive Medical Image Computing Workshop, 2014.
    Best paper award for impact and usability.

Nerve Bundle Tracking

Mapping of nerve bundles is essential for diagnosis and treatment planning of spinal pathologies, and offers substantial benefits for image-guided interventions. Manual segmentation, however, is quite challenging and time consuming, and automatic segmentation cannot use the traditional methods due to the challenging image and nerve properties. In this work we present a more general representation and particle filter based model for nerve segmentation with minimal input from a radiologist.

Sediment Redistribution in Sea Level Theory

Correct predictions of post-glacial sea level change have been instrumental in advancing our understanding of Earth’s internal structure, ice age and modern climate, and human migrations. In this work, we introduced a model for sediment redistribution built on top of a (modified) sea-level theory, and showed the improvements of the new model in an area with heavy sediments.

Genetic Variation Detection Framework

With VARiD, an HMM-based model and tool for simultaneously calling genetic variants from multiple platforms, we showed a significant advantage of modelling color-space and (multiple) letter-space reads into one framework.

Genome Variation Discovery with HTS (Review)

This was a review examining the algorithms behind rapid recent developments in genome variation with high througput sequencing.

Read Mapping

With Marc Fiume, I worked on the statistical part of Stephen Rumble's SHRiMP - the popular read mapping algorithm which could handle both letter-space and color-space technologies simultaneously. SHRiMP has significantly evolved since the original publication.

Sequence Alignment with Flexible Distance Measures

FRESCO is an alignment algorithm that can align sequences given a wide range of distance functions, allowing sequence alignment with more complex models or very particular settings.

Side Projects

Boston Timescape Dataset

Since 2010, I've taken pictures of the Boston Skyline from a high vantage point. The more than 400,000 images have been captured with different high resolution cameras (SLRs, GoPros, P&S, cell phone cameras). Some pictures are sin\ gletons (just one picture was taken for, say, that week or that day) whereas others are part of series or timelapses (and vary from 1/second to 1/minute, etc). I'm happy to share this data and am currently setting up a repository for it\ . There are several cool projects that could be done. Go to the project homepage to explore samples from the data with tipiX, and get more information!


SciEx is an initiative started by Zoya Bylinskii and myself to promote excitement in science. We are starting with a video competition where entries will aim to be stimulate interest in science for the younger generation in a way that x-games youtube videos stimulate interest in sports.

Boston Ballet Timescape

A video project with David Gifford and Boston Ballet. We want to capture the movement of the ballet at various time and personal scales, and here present a timelapse with various speeds and angles at practice in A Day of Grace with Boston Ballet.

K12 Video - Machine Learning

MIT+K12 is a project that offers short video tutorials that help educate K-12 students about various science and engineering topics. Together with a few friends we are working on several videos about Machine Learning, Probability and Algorithms. Once they are finished and (hopefully) accepted by the MIT+K12 group, they should be viewable on their website and YouTube.

Project Ideas for Computer Vision

Over time I've gathered several research directions/projects (mostly relevant to computer vision) that I wished I could do myself but realized they don't fit in my schedule. Some are small exercises (a good project course) and others are research directions. Talking to many people, I realized others are in this scenario - i'm working on introducing a site for sharing these ideas so that others might do it.

Space Cam

Together with Michael Rubinstein we launched a DYI balloon up to the lower atmosphere and retreived video from two GoPro hero 2, accelerometer and and GPS data.

Other Write-ups

Original design: MiniFolio