Deep Learning-Based Nuclear Lobe Count Method for Differential Count of Neutrophils

Area of Science:

• Hematology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Neutrophil nuclear lobulation is crucial for diagnosing hematological disorders like sepsis.
• Current manual assessment is time-consuming, prone to errors, and lacks standardization.
• Sepsis is associated with fewer nuclear-lobed and stab-formed neutrophils, indicating potential for early diagnosis.

Purpose of the Study:

• To develop and pilot a deep learning system for automated neutrophil differentiation based on nuclear lobulation.
• To assess the accuracy of deep learning in classifying neutrophil nuclear segmentation.

Main Methods:

• Utilized deep learning with a novel convolutional neural network architecture (Sony Neural Network Console).
• Processed 600 digital images of neutrophils stained with May-Grünwald Giemsa stain.
• Employed the Cellavision DM-96 automated digital microscope for image acquisition.

Main Results:

• The deep learning system achieved up to 99% accuracy in differentiating neutrophils into four groups based on nuclear segmentation.
• Successfully classified band-formed, two-, three-, and four- to five-segmented neutrophils.

Conclusions:

• Deep learning offers a highly accurate and automated approach for neutrophil differentiation by nuclear segmentation.
• This technology shows promise for improving the efficiency and reliability of diagnosing hematological conditions, including sepsis.