Feature extraction using traditional image processing and convolutional neural network methods to classify white...

Area of Science:

• Hematology
• Medical Image Analysis
• Computational Biology

Background:

• White blood cells (WBCs) are crucial indicators of health, with alterations signaling hematological disorders.
• Manual microscopic evaluation of WBCs is standard but subjective, relying on expert skill.
• Automated classification methods offer potential for objective and efficient analysis.

Purpose of the Study:

• To compare the efficacy of traditional image processing versus a deep learning approach (AlexNet) for WBC classification.
• To evaluate feature extraction methods for automated WBC analysis.
• To assess the diagnostic potential of computational methods in identifying hematological disorders.

Main Methods:

• Feature extraction using traditional image processing techniques.
• Feature extraction utilizing AlexNet, a pre-trained convolutional neural network.
• Classification of WBCs using a neural network with both feature sets.

Main Results:

• The convolutional neural network approach yielded slightly superior classification results compared to traditional methods.
• Both methods achieved high average accuracy and sensitivity, reaching 99% for WBC classification.
• The study validates the effectiveness of automated feature extraction for hematological analysis.

Conclusions:

• Both traditional image processing and deep learning (AlexNet) are viable for automated WBC classification.
• Deep learning methods offer a promising avenue for enhancing the accuracy and objectivity of hematological disorder detection.
• Method selection can be guided by data availability and resource constraints.