Basic Information

  • Project Title / Domain: Ultrasound Image Segmentation
  • Name: Anoushka Dey
  • Guide: Siddharth Tallur

Description of Abstract / Project

Work involved using UNets for ultrasound image segmentation along with the use of self supervised learning models such as the Barlow Twin model. The first half of the project was spent in literature review and exploratory work. This was followed by the creation of the required UNet, followed by the implementation of the Barlow Twin model and its subsequent integration in the UNet.

Whom did you work with?

The prof directly

Relevant Courses

Fundamentals of Digital Image Processing (CS 663 Autumn 2023-2024) , Advanced Image Processing (CS 754 Spring 2023-2024), Introduction to ML (CS 419 Spring 2023-2024)

How did you connect with the prof?

I mailed him and requested a meet in May itself. We had a discussion based on my research interests and he gave me a broad topic for exploration

Expectations from Prof

The prof expected me to balance time between exploratory work and implementation work. He set up regular meets to discuss weekly goals and milestones. As per my expectation, the prof was really systematic and enthusiastic about the project.

Logistics of the Project

He set up regular weekly meets to discuss the progress of the project. We planned and created an approximate milestone chart as to what our future weekly meetings would entail. Before the midsem, it was mostly exploratory and hence fixed milestones were not enforced. After the midsem, much more structured goals were required to be achieved. The prof could also be reached easily via Teams and I could have meets if I felt that a particular topic needed his attention immediately.

Reading Material

The first few papers were provided by the prof. I researched and found out more papers and articles which matched my project topic.

Tools / Simulations / Softwares / Hardwares Used

Python

Outcomes / Deliverables

Ensure the bring up and implementation of the UNet and the Barlow Twin model and use it to perform ultrasound image segmentation on a set of medical images

Extension to DDP

I did not extend it to DDP. I had taken the course EE618 CMOS Analog VLSI Design. As a result, I became interested in the field of circuits and hence in order to explore it, I decided to pursue my DDP under Prof. Rajesh Zele.

Type of Project

implementation-driven

Connection to DDP

nan

Advice

Be regular with your work and even if some weeks are hectic and you aren’t able to do as much, ensure that you inform the prof about your constraints and work load that week. He will appropriately scale his level of expectation for the meet. Ensure that you attend and meet up with the prof regularly and are also in constant contact via Teams. Prof. Satish Mulleti does research in these areas as well.