medical imaging dataset collection platform
Building India’s Medical Imaging Datasets
Role
Product designer
Platform
Webapp
Tool stack
Figma, Airtable, Jira
Project Duration
3 Months
MIDAS is a platform for efficient access and navigation of medical data. Tailored for healthcare professionals and researchers, the site not only provides an intuitive interface for exploring comprehensive datasets but also serves as a valuable resource for AI, facilitating learning and analysis to identify diseases in real-world scenarios.
Background
Health AI has the potential to revolutionize global healthcare by enhancing early disease detection, diagnosis accuracy, personalized treatment plans, and hospital workflows. Despite numerous Health AI solutions for conditions like COVID-19, Diabetic Retinopathy, and Tuberculosis, their real-world effectiveness has not fully met expectations.
One way to improve these solutions is by training and validating them with Indian gold-standard datasets. However, India's fragmented health system makes it difficult to create high-quality, accessible datasets.
To solve this, we are developing an open system using a hub-and-spoke model to provide appropriate access to training and validation datasets of the Indian population. This Digital Public Good Infrastructure is being built in collaboration with key stakeholders like ICMR, ABDM, and Niti Aayog, with research support from IISc Bengaluru.
In healthcare, the early detection of diseases using AI models has great potential to improve patient outcomes. However, developing these AI models depends heavily on having access to high-quality, well-organised healthcare datasets.
Right now, India doesn't have a centralised place to store healthcare datasets. (datasets stored locally, Like MRI,CT SCAN, XRAYS, etc.), making it hard for researchers and developers to access and use them effectively.
To solve this issue, we need to create a platform where users can upload their healthcare datasets. This platform should also have a system to verify the quality and reliability of these datasets.
As we started deep diving into the research insights and after endless meetings with doctors, we came to the conclusion that we need a platform for multiple users under one umbrella. Following are the users that are required to produce gold standard healthcare data.
End users
Who consumes the data that are published. and can be use this data to run their ML model.
Platform Admin
MIDAS (ICMR) who manages the entire platform - created the projects and assigns the Hubassigned
HUB
These people are assigned by the ICMR to be a manager of the projects these users will be regional hospitals who manages the spokes that are responsible for uploading the data to the platform
Spoke
Spokes are the third in the hierarchy which collect the data from patients and upload into the platform.
Annotators
An annotator carefully reviews and labels medical data to ensure it's accurate and high-quality.
Adjudicators
When the conflicts happen with the annotation. Adjudicators comes in place to make sure annotation decision are not biased and security check for the annotation.
Curators
These people are from the directory board of the platform(mostly ICMR doctors) to ensure the quality of the datasets
As a starting point, I figured out what are the key areas I need to focus on and listed down the requirements for it.
Bunch of questions need to be answered, Like
Who are the competitors world wide?
What kind of data will be uploaded? what format is it?
How it will be travelled?
What will be the expected size of the dataset?
Will user able to upload the single file? or batch?
and many more
after discussion with team and site co-ordinators I got up my answers
I quickly started with the process to make a User stories and information architecture
It is too obvious that non-tech people use the platform. I also need to ensure that all processes are shown upfront, so users can access everything in one place or one click ahead. The UX writing needs to be understandable for users, using medical terms.
So, after designing the flows and journeys, I started with paper sketches to define the page layout. We must has to create a framework that will be used across the board.
After the discussion and loop feedback, we have adopted a carbon design system, In order to achieve our requirements we have made a significant changes in the design system.
After Mid-fi designs presentation with clients we proceed to Hi-fi designs.
After feedback and Iterations we have managed to complete all the screens with documentation and handed over to tech team.
The site is live for the end consumerhttps://midas.iisc.ac.in/landing-pageNow, People can run and test there ML model’s with gold standard data.
After the discussion and loop feedback, we have adopted a carbon design system, In order to achieve our requirements we have made a significant changes in the design system.
After Mid-fi designs presentation with clients we proceed to Hi-fi designs.
After feedback and Iterations we have managed to complete all the screens with documentation and handed over to tech team.
The site is live for the end consumerhttps://midas.iisc.ac.in/landing-pageNow, People can run and test there ML model’s with gold standard data.
We managed tight deadlines like champs, delivering the project ahead of schedule with the small but mighty design team. From this point, I learned that team size is not important when we have the passion to drive forward.unimportant
We figured out a bunch of new stuff on the go, especially with medical terms, proving that learning on the job is a superpower.
We made a lot of assumptions as our starting point, turning them into gold through constant tweaking and improvement based on feedback from doctors.
Went straight to the source—and talked to the doctors to unravel what they needed, turning their vague ideas into a design that fits like a glove
After the first version release, Midas received significant recognition from AIIMS (Delhi), ICMR, and IISc (Bengaluru). It was also presented at the G20 summit for AI and approved by Niti Aayog.We figured out a bunch of new stuff on the go, especially with medical terms, proving that learning on the job is a superpower.