GSoC Week 6

This week, I was deeply involved in refining and optimizing the model selection hook's implementation on the AI model service end. I also discovered that the existing cheXnet model was not integrated correctly, and I spent some time rectifying that. On top of these tasks, I'm concerned that I have not been able to arrange a meeting with my mentor despite numerous requests. Hopefully, that will change next week.


Refining the Model Selection Method

Last week, I implemented the model selection method on the AI Model Service end, but some settings, such as the DICOM tag information required for model selection, were hardcoded. Other property-related information was also hardcoded. This week, I focused on separating the hardcoded elements and parameters into a separate YAML configuration file and refactoring the codebase. I learned that using a YAML configuration simplifies the maintenance of the codebase by reducing boilerplate code.


Subsequently, I updated the model call method to be dynamic so that task information or model information could be passed dynamically to the bounding-box method.


In its previous iteration, the bounding box method could only call the cheXnet model, as it was the only model available. The requirement now is to decouple the bounding box method to enable it to call multiple models such as cheXnet, mammogram, brain, etc.


Previously, the call would be:


https://localhost:5000/bounding-box


Now, it's been updated to:


https://localhost:5000/bounding-box/<task>


Where <task> can be cheXnet, Mammography, etc.


Fixing the cheXnet Model

While testing the cheXnet model, I discovered that the model weights were not loading correctly or were incompatible with the Densenet121 model present in the codebase. I determined that this was because the model weights had been trained on an older PyTorch version that was no longer available or compatible. I faced many issues because the model weights were sourced from a git repo codebase that did not specify the package versions.


To rectify this, I acquired a codebase for the cheXnet model and trained it. After properly evaluating the new model for AUC-ROC scores, I used the newly trained model weights to integrate it into the application. I also integrated it with newer grad cam implementations and libraries, which took some time but ultimately worked. Now, the cheXnet model functions as intended.


I have not yet submitted a Merge Request for this work since I need to review the updates with my mentor.


Addressing Failing Test Cases in the AI Model Service

After pushing the updated code for the model selection and the cheXnet model, I encountered some failing test cases. These failures resulted from changes in the bounding box method. After updating the test cases, they ran without any issues.


However, there's still one failed test case for the newly integrated cheXnet model. It fails when DICOM images are passed as parameters, but it works fine when JPEGs are used. I suspect there might be an issue with data loading for DICOM, and I will be working on this.

Comments

Post a Comment

Popular posts from this blog

Google Summer of Code 2023: AI Model Service Auto Selection by Modality - Final Report

Image
Introduction GSoC is one amazing program that lets students and professionals get into the world of open source, contribute code and also learn back from the program. I definitely learnt a lot by contributing to my mentoring organization LibreHealth through GSoC'23. I now understand the value of open source and will become a lifelong contributor wherever and whenever I can. LibreHealth is an umbrella organization where multiple health information technology projects collaborate. I was actively involved in contributing code to their LibreHealth Radiology application  through the project "AI Model Service Auto Selection by Modality". My contributions to the application were more involved into implementing AI models in the application and also building modules that do automatic model selection, example, when a Radiologist opens up a Mammography image the application automatically triggers the Mammography AI model to produce inferences on the image. Objectives The project des
Read more

My First GSoC - LibreHealth Community Bonding Period

GSoC Acceptance I was amazed to know that I got accepted into Google Summer of Code 2023. This is my first time applying and getting into GSoC. I remember the time when I sat at the library for hours, working on the preliminary tasks, figuring out solutions, and writing an extensive proposal. Reiterating through the solution and proposal back and forth, I watched lots of YouTube videos and read blogs providing tips to draft a good proposal. All the efforts paid off when I received the email of acceptance. The Org & Project I drafted a proposal to the LibreHealth Organization. LibreHealth is collaborative community for free & open source software projects in Health IT, and is a member project of Software Freedom Conservancy. It is amazing to just think about how contributors from all over the world come together and build things that impact the world positively. This idea is what brought me into open source. Before jumping into the project details, here are a few things about m
Read more

GSoC Week 2

Image
By the end of last week, I was able to meet my mentor, which helped me gain direction for the project. We discussed details and challenges regarding the implementation of the AI model selection based on image modality. Although Java is a new language to me, I wanted to learn and work my way through it for the project. This week, my task was to understand parts of the LibreHealth toolkit code and prepare a list of tasks at an atomic level. After that, I will need to update them to the GitLab issues list. Coming up with the task list To come up with the task list, during the meeting with my mentor, I received some guidance on where to start with code exploration and understanding. Initially, it was challenging for me to understand the code, so I started looking at Java tutorials on YouTube. I gained some understanding of how Java works. However, I knew this was not enough for me to completely understand the application, and it would take time, so I also started looking for patterns in th
Read more