Leukocytes Classification for Leukemia Detection Using Quantum Inspired Deep Feature Selection.

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence

Background:

• Leukemia diagnosis relies on accurate classification of white blood cell (WBC) subtypes.
• Automated WBC classification using deep learning shows promise but faces computational challenges due to large feature sets.
• Dimensionality reduction is crucial for improving model performance and reducing complexity in WBC classification.

Purpose of the Study:

• To develop an improved pipeline for WBC subtype classification.
• To enhance model performance and reduce computational complexity through intelligent feature selection.
• To leverage transfer learning and quantum-inspired evolutionary algorithms for accurate and efficient WBC classification.

Main Methods:

• Utilized transfer learning with deep neural networks for feature extraction from WBC images.
• Implemented a customized quantum-inspired evolutionary algorithm (QIEA) for wrapper feature selection.
• Classified reduced feature vectors using multiple baseline classifiers on a public dataset of 5000 WBC images.

Main Results:

• Achieved a classification accuracy of approximately 99% for WBC subtypes.
• Reduced the feature vector size by 90%, significantly decreasing computational complexity.
• Demonstrated superior convergence performance compared to classical genetic algorithms.

Conclusions:

• The proposed pipeline effectively classifies WBC subtypes with high accuracy and reduced computational cost.
• Quantum-inspired evolutionary algorithms offer an efficient approach for feature selection in medical image analysis.
• This methodology holds significant potential for improving leukemia diagnosis through automated WBC classification.