Starting My GSoC Journey with DIPY#

Hello everyone!

My name is Tomás Guija Valiente, and I am excited to share that I have been selected as a Google Summer of Code contributor with DIPY under the Python Software Foundation. Over the next few months, I will be working on improving DIPY’s image registration tools by adding new similarity metrics, with a special focus on Mutual Information for deformable registration.

This first post is a short introduction to me, the project, and what I have been doing during the Community Bonding period.

About Me#

I am a Computer Vision Researcher in Medical Imaging at PROMISE - LAIMBIO, the medical imaging lab at Universidad Rey Juan Carlos in Madrid, Spain. Before that, I studied at Universidad de Sevilla, where I completed a double Bachelor’s Degree in Mathematics and Computer Science, and later earned a Master’s Degree in Artificial Intelligence at Universidad Politecnica de Madrid.

During my Master’s, I became increasingly interested in open source and research. That eventually led me to my current work, where I focus on deep learning techniques for medical image processing and generative methods. GSoC felt like a very natural next step: it combines open source, medical imaging, scientific software, and the chance to learn from a strong technical community.

My selected project is titled **Adding New Similarity Metrics for DIPY’s Image Registration Frameworks** (see full details here). It is planned as a short GSoC project, with an initial scope of 90 hours and an expected end date near the beginning of August. If needed, the project scope may be expanded as the work progresses.

Community Bonding Period#

The Community Bonding period has been a great way to enter the project step by step. These first weeks were dedicated to meeting mentors, understanding the project goals more clearly, getting familiar with DIPY’s development workflow, and turning the initial proposal into a more concrete implementation plan.

The first meetings were a warm welcome for both me and my colleague Medha, who will be contributing to FURY, a closely related project. We had time to introduce ourselves, discuss expectations, ask questions, and of course start looking at code.

Project Overview#

The main goal of my project is to implement a new similarity metric for deformable medical image registration in DIPY. In particular, I will work on Mutual Information, a metric that is especially useful for multimodal registration problems.

DIPY already includes a Mutual Information implementation for affine registration. However, deformable registration currently relies on other metrics. Adding Mutual Information to this part of the registration framework could make DIPY more flexible for real medical imaging workflows, where images from different modalities often need to be aligned accurately.

Alongside the main implementation, I also plan to explore a few secondary goals:

  • Build a fair benchmark for comparing DIPY’s registration framework against other well-known open source solutions.

  • Use those comparisons to better understand the strengths, limitations, and possible improvements of the current optimization strategies.

  • Continue working, if time allows, on my existing pull request: Cython implementation for the Python nested triple loop on localpca eig path.

Progress So Far#

During the Community Bonding period, I started by studying the existing registration code and reading more about how DIPY currently structures its affine and deformable registration tools. I also began setting up a benchmarking workflow for SyN registration comparisons, which should help evaluate future changes in a more systematic way.

In parallel, I have been thinking through the design of the Mutual Information implementation. One of the important questions is how to make the metric cleanly integrated with the current codebase while keeping the solution general enough to be useful beyond a single registration path.

What’s Next?#

As the Coding period begins, my next steps are:

  • Finalize the implementation plan for Mutual Information in deformable registration.

  • Study existing approaches in DIPY and in other open source registration tools to identify useful design patterns.

  • Start implementing the metric in a way that fits naturally into DIPY’s current registration framework.

  • Add tests and examples so the new functionality is easier to review, maintain, and use.

  • Continue refining the benchmarking setup for fair comparisons.

First Impressions#

These first weeks have already made me feel welcome in the DIPY community. I am especially grateful to Serge and Atharva for their trust, guidance, and encouragement as I get started.

I am looking forward to learning a lot, contributing useful code, and sharing the process as the project develops. There is still plenty to discover, but that is exactly what makes the beginning of this journey so exciting.

Find Me Online#

Thank you for reading!

Google Summer of Code Final Work Product  
On this page