Medical Imaging & Diagnostics
What was the Problem?
Diagnosing retinal diseases accurately requires expensive medical equipment and highly trained specialists, making it inaccessible in remote areas.
How I Solved It
Developed a lightweight deep learning model trained on an 85K+ image dataset that accurately classifies retinal diseases on low-cost, resource-constrained hardware.
Effort Reduced
Automated the initial screening process, saving doctors approximately 40% of their time per patient diagnosis.
Profit / Impact Achieved
Reduced diagnostic equipment costs by over 70% for rural clinics, enabling wider deployment and early disease detection.
Difficulties Faced
The dataset was highly imbalanced and contained many noisy or poor-quality retinal images.
How I Overcame It
Implemented a strict data quality pipeline (removing 5% of noisy samples) which improved the model's accuracy more effectively than increasing model complexity.