BCNet: A Novel Network for Blood Cell Classification.

Area of Science:

• Medical image analysis
• Computational biology
• Artificial intelligence in healthcare

Background:

• Blood diseases like leukemia and anemia stem from various environmental factors and drug use.
• Accurate blood cell classification is vital for diagnosing and managing these conditions.
• Current classification methods, including manual feature extraction and deep learning, face challenges like tedium, computational cost, and overfitting on small datasets.

Purpose of the Study:

• To develop and evaluate novel, efficient models for automatic blood cell image classification.
• To address limitations of existing methods, particularly for small medical datasets.
• To propose an ensemble model that improves classification performance and robustness.

Main Methods:

• Seven novel blood cell classification models were proposed, including BCNet, utilizing a ResNet-18 backbone pre-trained on ImageNet.
• The final layers of ResNet-18 were replaced with three randomized neural networks (RNNs): RVFL, ELM, and SNN.
• BCNet employs an ensemble strategy, combining predictions from the RNNs via majority voting for final classification.

Main Results:

• The BCNet model demonstrated superior performance compared to other proposed models and state-of-the-art methods.
• Achieved high performance metrics: 96.78% accuracy, 97.07% average precision, 96.78% average F1-score, and 96.77% average recall.
• The ensemble approach effectively mitigated overfitting issues common with small medical datasets.

Conclusions:

• The developed BCNet model offers a significant advancement in automated blood cell classification.
• Its high accuracy and efficiency make it a promising tool for clinical diagnostics.
• The study highlights the potential of ensemble randomized neural networks for medical image analysis.