Examination of blood samples using deep learning and mobile microscopy.

Area of Science:

• Medical Diagnostics
• Computational Pathology
• Mobile Health Technology

Background:

• Microscopic analysis of blood is crucial for health assessment and disease diagnosis.
• Traditional laboratory blood tests are time-consuming and labor-intensive.
• Point-of-care systems offer potential for immediate bedside diagnostics.

Purpose of the Study:

• To develop and evaluate deep learning algorithms for blood cell detection using a mobile microscope.
• To assess the feasibility of a smartphone-based system for point-of-care blood analysis.
• To achieve accurate segmentation and classification of blood cells.

Main Methods:

• Human blood samples were visualized using a low-cost mobile microscope, ocular camera, and smartphone.
• Deep learning instance segmentation models (Mask R-CNN, Mask Scoring R-CNN, D2Det, YOLACT) were trained and optimized.
• Extensive modifications were made to adapt networks for detecting numerous small objects like blood cells.

Main Results:

• Instance segmentation models achieved accurate detection and classification of all blood cell types.
• Qualitative evaluation yielded a mean average precision of 0.57 and mean average recall of 0.61.
• Quantitative analysis demonstrated detection of 93% of ground truth blood cells.

Conclusions:

• Mobile blood testing with deep learning achieves diagnostic accuracy comparable to traditional methods.
• This approach enables rapid, cost-effective, and location-independent patient care.
• Future applications can significantly enhance healthcare accessibility, especially in remote or underserved areas.