Blood Cell Classification Based on Hyperspectral Imaging With Modulated Gabor and CNN.

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning for Healthcare

Background:

• Accurate cell classification, particularly white blood cells, is crucial for disease diagnosis and management.
• Hyperspectral imaging offers richer spatial and spectral information compared to traditional optical microscopy for cell recognition.

Purpose of the Study:

• To propose a novel blood cell classification framework, MGCNN, integrating modulated Gabor wavelets with deep convolutional neural network (CNN) kernels.
• To enhance the discriminative power of features learned from medical hyperspectral images for improved cell classification.

Main Methods:

• Developed a novel framework (MGCNN) combining modulated Gabor wavelets and deep CNN kernels for hyperspectral blood cell classification.
• Integrated multi-scale and orientation Gabor operators with CNN kernels, transforming kernel learning into the frequency domain.
• Utilized the frequency and orientation characteristics of Gabor wavelets to extract more representative and discriminative features.

Main Results:

• The proposed MGCNN model demonstrated superior classification performance compared to traditional CNNs and support vector machine approaches.
• Achieved enhanced classification accuracy, especially effective in small-sample size training situations.
• Validated the effectiveness of frequency-domain feature learning using modulated Gabor wavelets for hyperspectral cell analysis.

Conclusions:

• The MGCNN framework offers a significant advancement in hyperspectral blood cell classification.
• The integration of Gabor wavelets with CNNs provides a powerful approach for learning discriminative features in medical imaging.
• This method shows promise for improving diagnostic accuracy in resource-limited or small-sample clinical scenarios.