Automated screening of sickle cells using a smartphone-based microscope and deep learning

Area of Science:

• Medical Diagnostics
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Sickle cell disease (SCD) is a global health priority, particularly in resource-limited regions where diagnosis is inconsistent.
• High mortality rates in African children with SCD are often preventable with timely diagnosis and treatment.
• Current diagnostic methods can be inaccessible in many affected areas.

Purpose of the Study:

• To develop and validate a deep learning framework for automated sickle cell screening using smartphone microscopy.
• To create a cost-effective and accessible diagnostic tool for SCD in resource-limited settings.

Main Methods:

• A two-stage deep neural network framework was developed for image enhancement and cell segmentation.
• The first network standardizes smartphone-captured blood smear images to laboratory-grade quality.
• The second network performs semantic segmentation to differentiate between healthy and sickle cells.

Main Results:

• The mobile screening method achieved approximately 98% accuracy in detecting sickle cells.
• The system demonstrated a high area-under-the-curve of 0.998 in blind testing.
• The framework successfully identified sickle cells from blood smears of 96 unique patients.

Conclusions:

• The developed deep learning framework offers a highly accurate and cost-effective method for SCD screening.
• This mobile diagnostic tool has significant potential for widespread use in resource-limited settings.
• The technology could be adapted for screening other blood cell disorders.