Image Analysis Using Machine Learning for Automated Detection of Hemoglobin H Inclusions in Blood Smears - A Method...

Area of Science:

• Hematology
• Medical Imaging
• Artificial Intelligence

Background:

• Morphologic rare cell detection is labor-intensive and operator-dependent.
• Hemoglobin H (HbH) inclusions in red blood cells are key indicators for alpha-thalassemia screening.
• Artificial intelligence (AI) offers potential improvements for rare cell detection through image analysis.

Purpose of the Study:

• To develop a convolutional neural network (CNN)-based algorithm for detecting HbH inclusions in red blood cells.
• To evaluate the performance of the AI algorithm across various imaging conditions and platforms.
• To establish optimal parameters for AI-assisted red blood cell analysis in alpha-thalassemia trait.

Main Methods:

• Training and testing a CNN algorithm using digital images of HbH-positive and HbH-negative blood smears.
• Validating software performance on images from different magnifications and scanning platforms.
• Developing a red blood cell counting model and using Poisson modeling to determine minimum red cells required for analysis.

Main Results:

• The AI software achieved high sensitivity (approx. 91%) and specificity (approx. 99%) for HbH+ cell detection.
• AI-aided diagnosis demonstrated good inter-rater reliability and high slide-level classification accuracy on whole slide images (WSIs).
• Analysis of 2.4 x 10^6 red blood cells is estimated to minimize misclassification, validated on 78 image sets.

Conclusions:

• Whole slide image (WSI) analysis using AI is effective for morphologic rare cell detection.
• The developed AI software shows promise for clinical application in alpha-thalassemia diagnosis.
• Further development and clinical validation studies are recommended to compare AI-aided diagnosis with routine methods.