Deep learning-based label-free hematology analysis framework using optical diffraction tomography

Area of Science:

• Biomedical Engineering
• Computational Biology
• Medical Diagnostics

Background:

• Traditional hematology analysis relies on time-consuming and labor-intensive chemical staining processes.
• Label-free imaging presents a cost-effective and efficient alternative for hematology analysis.

Purpose of the Study:

• To develop a label-free hematology analysis framework using optical diffraction tomography and deep learning.
• To accurately detect and classify various blood cell types without chemical staining.

Main Methods:

• Utilized optical diffraction tomography for label-free imaging of blood cells.
• Employed the fully convolutional one-stage object detector (FCOS), a deep learning model, for cell detection and classification.
• Classified detected cells into four groups: red blood cells, abnormal red blood cells, platelets, and white blood cells.

Main Results:

• The object detection model achieved a mean average precision (mAP) of 0.977 for blood cell detection.
• Achieved high accuracy in four-class blood cell classification with a weighted F1 score of 0.9708 and total accuracy of 0.9712.
• Demonstrated reasonable correlation with reference equipment for Mean Corpuscular Volume (MCV) (0.905) and Mean Corpuscular Hemoglobin (MCH) (0.889).

Conclusions:

• The proposed framework successfully demonstrates label-free detection and classification of blood cells using optical diffraction tomography.
• This approach offers a promising, efficient, and cost-effective alternative to conventional hematology analysis.
• The study validates the potential of integrating advanced imaging and deep learning for improved diagnostic tools.