Human peripheral blood leukocyte classification method based on convolutional neural network and data augmentation.

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning

Background:

• Accurate classification of human peripheral blood leukocytes is crucial for diagnosing blood diseases.
• Convolutional Neural Networks (CNNs) show promise for automated microscopic leukocyte image classification.
• Dataset scarcity and imbalance often limit the accuracy of CNN-based classification methods.

Purpose of the Study:

• To improve the classification accuracy of microscopic leukocyte images using CNNs.
• To address challenges of dataset scarcity and imbalance in leukocyte image datasets.
• To develop an effective automated method for leukocyte classification in peripheral blood samples.

Main Methods:

• Designed a deep CNN model for microscopic leukocyte image classification.
• Proposed a novel data augmentation method based on feature concentration to enrich the dataset.
• Implemented a resampling technique to balance the representation of five leukocyte types within training batches.

Main Results:

• Achieved an average testing accuracy of 97.6% for leukocyte classification.
• Demonstrated classification precision above 93.4% and sensitivity above 92.5% for all five leukocyte types.
• Confirmed that both data augmentation and resampling methods significantly enhance classification accuracy.

Conclusions:

• A robust method for human peripheral blood leukocyte classification using CNNs and data augmentation was developed.
• The proposed data augmentation effectively resolves dataset scarcity issues.
• The resampling method successfully addresses dataset imbalance, leading to improved classification performance.