Related Experiment Video
Updated: Aug 20, 2025
Enumeration of Major Peripheral Blood Leukocyte Populations for Multicenter Clinical Trials Using a Whole Blood Phenotyping Assay
Published on: September 16, 2012
BCNet: A Deep Learning Computer-Aided Diagnosis Framework for Human Peripheral Blood Cell Identification.
Channabasava Chola1, Abdullah Y Muaad2, Md Belal Bin Heyat3,4,5
1Department of Electronics and Information Convergence Engineering, College of Electronics and Information, Kyung Hee University, Suwon-si 17104, Republic of Korea.
BCNet, an AI deep learning framework, accurately identifies eight blood cell types. It achieves 98.51% accuracy and outperforms other models in speed, enhancing healthcare diagnostics.
Area of Science:
- Hematology and Medical Diagnostics
- Artificial Intelligence in Healthcare
- Computational Biology and Bioinformatics
Background:
- Accurate identification of blood cells is crucial for assessing a patient's health status and managing infection risks.
- Manual blood cell identification is time-consuming and prone to human error, necessitating automated solutions.
- Deep learning (DL) offers potential for rapid and precise automated analysis of cellular images.
Purpose of the Study:
- To introduce BCNet, a novel artificial intelligence (AI)-based deep learning (DL) framework for automated blood cell identification.
- To evaluate the performance and efficiency of BCNet in classifying eight distinct types of blood cells.
- To compare BCNet's capabilities against existing state-of-the-art DL models for blood cell analysis.
Main Methods:
- Developed BCNet using a convolutional neural network (CNN) architecture with transfer learning capabilities.
- Conducted five-fold cross-validation experiments to assess model dependability and viability.
- Tested BCNet with ADAM, RMSprop (RMSP), and stochastic gradient descent (SGD) optimizers, comparing results with DenseNet, ResNet, Inception, and MobileNet.
Main Results:
- BCNet demonstrated superior classification performance with ADAM and RMSP optimizers, achieving a peak accuracy of 98.51% and an F1-score of 96.24% with RMSP.
- BCNet improved prediction accuracy by 1.94% (ADAM), 3.33% (RMSP), and 1.65% (SGD) compared to baseline models.
- BCNet exhibited significantly faster testing times per image (10.98, 4.26, 2.03, and 0.21 msec faster than DenseNet, ResNet, Inception, and MobileNet, respectively).
Conclusions:
- The BCNet framework provides a dependable and efficient AI-driven solution for automated blood cell identification.
- BCNet's high accuracy and rapid processing speed offer significant advantages over existing DL models.
- The proposed BCNet model shows strong potential for advancing diagnostic capabilities in healthcare facilities.
More Related Videos
07:58Quantitative Imaging of Lineage-specific Toll-like Receptor-mediated Signaling in Monocytes and Dendritic Cells from Small Samples of Human Blood
Published on: April 16, 2012
08:58Label-Free Identification of Lymphocyte Subtypes Using Three-Dimensional Quantitative Phase Imaging and Machine Learning
Published on: November 19, 2018
