Medical Imaging Diagnostics
CNN-based image classification for malaria diagnosis from blood-smear images, evaluated for generalization across datasets.
Diagnostic imaging models often score well on the dataset they were trained on and degrade on a different one collected under different conditions. This research trained and evaluated CNN-based image classification models across multiple blood-smear datasets, with the evaluation specifically designed around generalization rather than single-benchmark accuracy.
Primary technologies
- Python
- TensorFlow
- Keras
- CNNs
- Image classification
01 / Context
What this project is
Diagnostic imaging models often score well on the dataset they were trained on and degrade on a different one collected under different conditions. This research trained and evaluated CNN-based image classification models across multiple blood-smear datasets, with the evaluation specifically designed around generalization rather than single-benchmark accuracy.
Engineering problem
Blood-smear datasets vary in staining, imaging equipment, and collection conditions — a model that performs well on one can fail on another, which matters a great deal for a diagnostic tool.
Approach
Trained CNN-based classification models in TensorFlow and Keras and assessed how performance generalized across datasets rather than optimizing for one.
02 / Stack
What it's built with
Technologies
- Python
- TensorFlow
- Keras
- CNNs
- Image classification
03 / Outcome
What shipped
The findings were peer-reviewed and presented at ICDMIS 2025 (Springer) — see the Research section below.
Next step
Have something like this
you need built properly?
I work on the AI layer — retrieval, reasoning, and the plumbing underneath — and hand it over ready to integrate.