White Blood Cells Classification Using Entropy-Controlled Deep Features Optimization

Area of Science:

• Hematology
• Computational Biology
• Medical Imaging

Background:

• Accurate white blood cell (WBC) subtype identification is vital for diagnosing leukemia.
• Traditional manual blood smear analysis is time-consuming and prone to errors.
• Deep learning methods offer high accuracy but require extensive computational resources.

Purpose of the Study:

• To develop an efficient hybrid approach for WBC subtype classification.
• To reduce the computational cost associated with deep learning models for WBC analysis.
• To improve the accuracy and efficiency of automated WBC subtype identification.

Main Methods:

• Utilized transfer learning with DenseNet201 and Darknet53 for deep feature extraction from enhanced WBC images.
• Applied an entropy-controlled marine predator algorithm (ECMPA) for feature selection and reduction.
• Classified the reduced feature set using multiple baseline classifiers.

Main Results:

• Achieved an overall average accuracy of 99.9% on a dataset of 5000 synthetic WBC images.
• Reduced the feature vector size by over 95%, enhancing computational efficiency.
• Demonstrated superior convergence performance compared to traditional meta-heuristic algorithms.

Conclusions:

• The proposed hybrid method offers a highly accurate and efficient solution for WBC subtype classification.
• This approach effectively addresses the computational challenges of deep learning in medical image analysis.
• The ECMPA-based feature selection significantly optimizes the classification process for leukemia diagnosis.