Classification of white blood cells using capsule networks

Area of Science:

• Medical Imaging Analysis
• Computational Biology
• Artificial Intelligence in Healthcare

Background:

• Accurate white blood cell (WBC) classification is crucial for disease diagnosis.
• Manual classification is slow and expertise-dependent.
• Limited medical data and preprocessing challenges hinder traditional machine learning and deep learning models.

Purpose of the Study:

• To evaluate capsule networks for WBC classification using small datasets.
• To address data limitations and preprocessing requirements in medical image analysis.
• To compare capsule network performance against established deep learning methods.

Main Methods:

• Classification of WBCs into five categories using capsule networks.
• Application of various model improvement techniques.
• Comparative analysis with prominent deep learning models like Convolutional Neural Networks (CNNs) and Transfer Learning (TL).

Main Results:

• Capsule networks achieved high accuracy (96.86%) on test data, overcoming overfitting issues.
• The proposed model demonstrated superior performance compared to CNN and TL models.
• Successful classification was achieved without extensive data preprocessing or augmentation.

Conclusions:

• Capsule networks are a viable and successful alternative for deep learning in medical data analysis, especially with limited sample sizes.
• The study highlights the potential of capsule networks to overcome common challenges in medical image classification.
• This approach offers a promising direction for developing efficient computer-aided diagnostic systems.