ReRNet: A Deep Learning Network for Classifying Blood Cells.

Area of Science:

• Medical diagnostics
• Computational biology
• Artificial intelligence in healthcare

Background:

• Accurate blood cell classification is crucial for disease detection and diagnosis.
• Current classification methods face limitations in accuracy and consistency, leading to potential diagnostic errors and increased workload for medical professionals.
• Automated classification systems can aid clinicians by providing objective diagnostic criteria, improving efficiency, and reducing human error.

Purpose of the Study:

• To develop and validate an automated blood cell classification network.
• To improve the accuracy and reliability of blood cell classification for disease diagnosis.
• To provide a robust tool for clinicians to assess disease type and severity.

Main Methods:

• A ResNet50-based ensemble of randomized neural networks (ReRNet) was proposed for blood cell classification.
• ResNet50 was utilized as the backbone for feature extraction.
• The extracted features were processed by three randomized neural networks (Schmidt neural network, extreme learning machine, and dRVFL), with results combined via majority voting.

Main Results:

• The proposed ReRNet achieved an average accuracy of 99.97%, average sensitivity of 99.96%, average precision of 99.98%, and average F1-score of 99.97%.
• The ReRNet demonstrated superior performance compared to four other state-of-the-art methods.
• The network's effectiveness was validated using 5x5-fold cross-validation.

Conclusions:

• The ReRNet is a highly effective method for automated blood cell classification.
• The proposed model offers significant improvements in classification performance, aiding in more accurate disease diagnosis.
• This approach has the potential to enhance clinical decision-making and patient care.