Updated 8/15/2025

UROP Project: Image Processing Platform for Coronary OCT Image Analysis

Our group has developed a variety of tools that quantify and analyze coronary optical coherence tomography images. However, our group would like to combine these projects into one single multipurpose platform that can be continued in through the future. As more tools are developed, it is important for our proposed platform to be scalable and allow the integration and improvement of platform components. We are seeking 4 UROPs who would work together on this through the fall and spring semester.

Responsibilities: These selected UROPs will work closely with a PhD student and with each other to develop and improve the unified image processing platform.

• UROP 1: This UROP will be a 1st or 2nd year student responsible for the front and back-end software engineering of the image processing platform. They will communicate with the other three UROPs to readily integrate their projects into the platform. You will build upon an existing platform and handle all user-server interactions.
  • Requirements: Node.js, Python, Docker, Google Cloud, GIT CI/CD
• UROP 2: This UROP will be a 3rd or 4th year student responsible for the development and improvement of an existing coregistration tool along with a meshing tool. The coregistration tool semi-automatically rotationally and spatially aligns two sets of arteries before and after intervention. Further improvement can be made to make this fully automatic using advanced feature detection methods. Our meshing tool creates realistic 3D patient specific geometries using 2D segmentation masks, and it would be beneficial to add a UI frontend to interact with the meshing backend for simplicity of use.
  • Requirements: Python, Image Processing, UI/UX
• UROPs 3-4: These UROPs will be 4th year students responsible for the data augmentation and retraining of our existing arterial mask segmentation networks based on the nnU-Net architecture and the MMSeg toolbox. We already have existing code, but our dataset is quite limited. We want to improve the model performance by adding more image + segmentation pairs to training while tuning the hyperparameters or modifying the architecture. We would also like to augment the model’s capabilities and segment the device and lipid in the artery. 
  • Requirements: Familiar with deep learning (specifically Pytorch framework) and hyperparameter tuning.

Prospective candidates should contact ajay_m@mit.edu with a resume and name which subproject is of interest.