Inspiration
- Nearly every minute, a child under 5 dies of malaria. Many of these deaths are preventable and treatable. In 2022, there were 249 million malaria cases globally that led to 608,000 deaths in total. Of these deaths, 76 per cent were children under 5 years of age. This translates into a daily toll of over 1,000 children under age 5. - UNICEF Data, 2024
What it does
- PlasmoAI is a diagnostic tool that detects malaria parasites in blood smear images captured through a smartphone-microscope combo. The model is fully offline, runs in real-time, and is optimized to work on Android phones, making it a practical fit for under-resourced rural health centers.
How we built it
- Dataset: NIH Malaria Dataset.
- Model: Lightweight CNN using MobileNetV2, trained with transfer learning.
- Deployment: Converted to TensorFlow Lite (.tflite) and integrated with an Android proof-of-concept app.
- Inference: Runs directly on-device without internet or cloud access.
Challenges we ran into
- Balancing model size and inference speed on low-spec phones.
- Getting clear microscope images with cheap plastic clip-on lenses.
- Addressing real-world noise and artifacts in field data.
Accomplishments that we're proud of
What we learned
- Well prepared data is more important than a massive model.
- Designing for the African context means planning for offline use and low power.
- Field-level health innovation is possible with commodity hardware and open-source tools.
What's next for PlasmoAI
- Gather field data from African clinics to further fine-tune and validate.
- Partner with NGOs and ministries of health for pilot deployments.
- Expand the AI model to include other bloodborne diseases, sickle cell for example.