Medical Imaging & Deep Learning Ensemble Architecture
Computer vision research prototype combining Convolutional Networks and Vision Transformers (ViT) for retinal fundus analysis. Not medical diagnosis.
Project Architecture & Case Study
The Challenge & Scope
Medical imagery exhibits subtle contrast shifts, varied illumination, and microscopic lesions where a single monolithic model architecture often degrades across edge cases.
Technical Architecture & Solution
Built a PyTorch preprocessing pipeline with CLAHE and advanced geometric augmentations. Implemented an ensemble voting strategy bridging convolutional architectures (EfficientNet/ResNet) and Vision Transformers.
Results & Key Deliverables
Developed a modular computer vision research harness validating multi-architecture model consensus without making unverified clinical diagnosis claims.
About the Project
A deep learning exploration analyzing biological tissue and pathological lesion patterns in retinal fundus photography. Investigates ensemble inference combining distinct visual model architectures (CNNs and Vision Transformers). Strictly an academic computer vision exploration; not intended for clinical diagnosis or medical treatment.
Architected, designed, and engineered entirely by Ahmet Mert Yiğitbaşı.
Technologies
- PyTorch
- Python
- Vision Transformers (ViT)
- CNN Architectures
- OpenCV & Image Processing
- Model Ensemble Architecture