CASE STUDY

Crowdsourcing for eye disease diagnosis

The Brief

In work with the University of Surrey, we developed a crowdsourcing system that trains users to spot signs of eye disease in photographs. We built a user-friendly web browser tool in which untrained users were able to annotate retina fundus photos and eye CT scans, receiving scores and leaderboard positions for their accuracy.

Brief and project roadmap

Before beginning development, we worked closely with the University of Surrey – to develop a project roadmap with well-defined goals.

 

User interface development

Based on careful discussion with the client, we designed and implement a full user interface

 

Advanced development: algorithms, crowdsourcing, and gamification

This project allowed us to deploy a number of our most advanced specialist skills, from bespoke algorithm development to crowdsourcing technologies and the incorporation “serious” games, including social logins and leaderboards.

Development

The project involved both the full development of a crowdsourcing system and the careful design of a user interface for the application. The system allowed users to spot anomalies that might indicate eye disease in photographs and pictures of eye CT scans, and to flag these in the system itself. Meanwhile, gamification elements encouraged them to keep participating while training them to get better at spotting signs of eye disease.

 

 

  • User flow, wireframes and visual design

    We carefully designed the application using user flows, wireframes and visual designs.

  • Crowdsourcing system development

    We built, from the ground up, a fully integrated crowdsourcing system. This included user accounts, image management, training modules, and functionality for further data imports. For image analysis, we implemented web-based annotation tools to enable user-friendly image annotation for multiple eye diseases.

  •  Gamification and user management

    Alongside the development of social logins and training resources, we built leaderboards, levels, and an achievements system in order to encourage users to ‘play’ more effectively.

  • Bespoke algorithm development

    We developed bespoke algorithms to determine the accuracy of users’ image annotations. Accuracy was tested on an individual level against provided ‘ground truths’, and also in comparison with a weighted average of other users’ responses.

computer vision | crowdsourcing | gamification | iOS app design | iOS app development | machine learning | UX | web application development

Get in touch today and find out how we turn vision to reality.


Get in touch