Classification of red blood cells as normal, sickle, or other abnormal, using a single image analysis feature

Area of Science:

• Hematology
• Medical Imaging
• Computational Biology

Background:

• Sickle cell anemia lacks effective treatments, necessitating accurate methods for clinical status evaluation.
• Morphological cell counting is a key method, but human inspection is subjective and lacks reproducibility.
• There is a critical need for automated, objective, and rapid cell classification in sickle cell disease.

Purpose of the Study:

• To develop and validate a quantitative digital-image analysis method for classifying sickle cells.
• To compare the accuracy of automated cell classification against human expert classification.
• To identify key cellular features for distinguishing normal, sickle, and abnormal red blood cells.

Main Methods:

• Blood samples from sickle cell anemia (SS, SC) patients and normal (AA) volunteers were analyzed.
• Cells were stressed using nitrogen bubbling, and 150 (sickle) or 100 (normal) cells per specimen were analyzed.
• A high-resolution image-analysis instrument extracted 42 features; recursive partitioning identified 'Form Factor' as the key feature for classification.

Main Results:

• Automated classification achieved high agreement with human experts: 89% (normal), 73% (abnormal), and 92% (sickle) for SS/SC samples.
• Agreement for normal (AA) samples was 92% (normal) and 76% (abnormal).
• Automated classification performance met or exceeded inter-observer agreement among human experts.

Conclusions:

• Quantitative digital-image analysis, particularly using the 'Form Factor', provides an accurate and reproducible method for sickle cell classification.
• This automated approach surpasses human expert reliability for evaluating sickle cell morphology.
• The technology holds promise for improving the assessment of sickle cell disease status.