An Eye Into the Heart

We are 2 high school students who are dedicated to creating innovative solutions that improve the quality of life for people around the world. Our focus is on developing cutting-edge technologies that address some of the most pressing challenges facing society today, from healthcare and education to sustainability and social impact. We believe that by leveraging the power of technology and design, we can create meaningful change and make a positive impact on the world.

This AI model relies heavily on multi-level geometric structure and pixel-level segmentation.

First the video’s frames are isolated and landmark-based Homography Estimation Models are used to re-centre the eye based on canthi (corners of the eye), brow and nose apex to centre the eye for following segmentation.

After that Semantic and Instance Segmentation is needed to isolate the eye via encoder decoder neuro-networks that allow for pixel perfect isolation of the image for diagnostic modelling. This process is also aided by Data Augmentation by randomly shifting brightness, contrast, and adding artificial shadows to the training images so the model learns to look at texture boundaries rather than light levels. This allows for the final model to be ‘tuned’ to the same condition for AI diagnostics.

Our diagnostic AI model relies on the low energy and easy to train and update K-Nearest Neighbour classification algorithm for the initial tier of risk stratification.

KNN was chosen not just because for the low cost and ease of training, they are an intuitive instance based learning algorithm that allows for constant update of the ‘vector’ base. By not relying on complex, highly resource-intensive mathematical models, it instead classifies a patient’s current cardiovascular health status based on how closely their data matches known historical patient profiles.

When the patient prefroms a scan, the app plots their data set (normalised image, demographic and Social Determinants of Health) into a new point as a vector in the feature. space.

If you want to have a go at our design, please click this link below

Here