Drop an image here, or click to choose one.
JPG or PNG of a single stained cell works best.
How this number is computed
The image is resized to 50×50 and shown to the model in 8 geometric views (rotations and flips). The reported probability is the mean of the 8 outputs; the interval is mean ± 1.96×SD of those 8 outputs. Wide intervals mean the model is unsure about that cell.
| View | P(infected) |
|---|
Classify in the cloud
Same model, same 8-view augmentation, run on a server instead of
your browser. Your image is sent to /api/predict and classified there.
About malaria
Malaria is an acute febrile illness caused by Plasmodium parasites, which are spread to people through the bites of infected female Anopheles mosquitoes. There are 5 parasite species that cause malaria in humans, and 2 of these species, P. falciparum and P. vivax, pose the greatest threat. P. falciparum is the deadliest malaria parasite and the most prevalent on the African continent. P. vivax is the dominant malaria parasite in most countries outside of sub-Saharan Africa.
Symptoms
- Fever
- Chills
- Headache
- Sweats
- Nausea
- Vomiting
- Body aches
- Fatigue
- Cough
How staging works
When a cell is flagged as infected, the pipeline outlines the parasite with contour detection and measures the ratio of parasite area to cell area. That ratio places the infection in one of three stages:
| Parasite-to-cell area ratio | Stage |
|---|---|
| 7% and above | Advanced schizont stage |
| 2.5% to 7% | Trophozoite stage |
| Below 2.5% | Initial ring stage |
Methodology from the original analysis pipeline: contrast adjustment and cleanup of the cell image, then OpenCV contour detection to outline the parasite and measure its area against the cell.
Malaria around the world
Reported cases and deaths by country, 2000 to 2017.
Countries with the most cases in 2017
Source: World Health Organization.
About this model
A compact CNN trained on about 27,000 NIH thin-blood-smear cell images (half parasitized, half uninfected). It scores 93.2% accuracy (95% CI 92.9–93.5), sensitivity 92.9%, specificity 93.5%, AUC 0.958 on those same images — the original train/test split was never recorded, so treat these as an optimistic upper bound, not a held-out test score. Mean TTA spread across 8 views is ±0.055 in probability. Full write-up, code, and figures: github.com/nepalanurag/Detection-of-Malaria-Using-CNN.
Research demo, not a medical device. Predictions can be wrong; see the error analysis in the repo before trusting it.