Automatic Image Segmentation of Monocytes and Index Computation Using Deep Learning.

Area of Science:

• Hematology
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• Accurate white blood cell classification is crucial for medical diagnosis, identifying various blood-related conditions.
• The Monocyte Index (MI) is vital for assessing transfusion compatibility, as monocytes can interact with red blood cells.
• Manual blood cell counting is time-consuming and error-prone, necessitating automated solutions.

Purpose of the Study:

• To develop an automated method for calculating the Monocyte Index (MI).
• To assess the feasibility of using deep learning for pixel-level segmentation of monocytes and red blood cells for MI calculation.
• To improve the efficiency and accuracy of blood analysis for transfusion decisions.

Main Methods:

• A custom dataset of microscope images was collected and annotated.
• The Mask R-CNN deep neural network model was trained for automatic pixel-level segmentation.
• COCO pre-trained weights were utilized to initialize the Mask R-CNN model.

Main Results:

• The Mask R-CNN model achieved 72% accuracy in automatic segmentation.
• The automated system demonstrated the ability to process large datasets rapidly and efficiently.
• The model's accuracy was comparable to that of a human medical laboratory scientist.

Conclusions:

• Deep learning, specifically Mask R-CNN, shows promise for automating Monocyte Index calculation.
• This automated approach can significantly reduce laboratory workload and enhance diagnostic efficiency.
• The system provides a valuable tool for clinicians in identifying suitable blood transfusion candidates.