Patient-level performance evaluation of a smartphone-based malaria diagnostic application.

Area of Science:

• Medical Diagnostics
• Parasitology
• Mobile Health (mHealth)

Background:

• Microscopy is standard for malaria diagnosis but limited by expert availability and time.
• Automated, machine learning-based systems offer potential solutions to improve malaria diagnosis.
• This study evaluates Malaria Screener, a smartphone application for malaria diagnosis.

Purpose of the Study:

• To evaluate the diagnostic performance of the Malaria Screener smartphone application.
• To compare Malaria Screener's accuracy against expert microscopy and PCR.
• To assess a new algorithm, PlasmodiumVF-Net, in post-study experiments.

Main Methods:

• 190 patients were recruited in rural Sudan.
• Malaria Screener screened Giemsa-stained blood smears.
• Expert microscopy and nested PCR served as reference standards; PlasmodiumVF-Net was also evaluated.

Main Results:

• Malaria Screener achieved 74.1% accuracy vs. expert microscopy and 71.8% vs. PCR (WHO Level 3).
• Post-study, Malaria Screener reached 91.8% accuracy (WHO Level 1) using a different calculation method.
• PlasmodiumVF-Net achieved 83.1% accuracy vs. microscopy and 81.0% vs. PCR (WHO Level 2).

Conclusions:

• Malaria Screener demonstrates potential for routine malaria screening in resource-limited areas.
• It is the first smartphone-based malaria diagnostic system evaluated at the patient-level in a field setting.
• Results provide a reference for future studies on mobile health malaria diagnostics.