Eyecare Project - OpthaDetect

Aside from my academic endeavours in teaching and researching ICT4D, I am also committed to developing projects that can solve problems beyond the classroom. In the second year of my PhD, I was selected as a year-long fellow of Cambridge King's (College) Entrepreneurship lab, where I pitched a project to solve social media communication problems for universities and academic enterprises in Africa. The success of my pitch and fellowship has encouraged me to partner and collaborate to develop more projects that can solve societal needs.

​

In particular, I am working with a machine learning engineer, Rodiyah Oluwa, to develop solutions to track Diabetic Retinopathy (DR) faster, in both emerging and frontier markets. Globally, getting results from eye scans takes 1 to 7 days. Our project, OpthaDetect, specifically seeks to reduce the wait time for DR to seconds, providing a confidence level statement for all results. 

​

DR is a progressive complication of diabetes and remains one of the leading causes of preventable vision loss worldwide. Early and accurate detection is essential to improving patient outcomes, especially in low-resource settings where access to specialist ophthalmic care is limited. OpthaDetect, therefore, applies deep learning for automated DR detection using high-resolution retinal fundus images.

​

A robust preprocessing pipeline was developed to enhance image quality and standardisation. The pipeline incorporated circular cropping, Fast Fourier Transform (FFT)-based denoising, and Contrast-Limited Adaptive Histogram Equalisation (CLAHE). Five deep learning architectures were implemented and evaluated for binary classification (DR vs. No DR), and model performance was assessed using accuracy, precision, recall, F1-score, and AUC-ROC. To strengthen interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to visualise regions influencing model decisions.

​

Our Minimum Viable Project can be accessed here. ​​