Related Experiment Video
Updated: Jul 7, 2025
08:19
Lensless On-chip Imaging of Cells Provides a New Tool for High-throughput Cell-Biology and Medical Diagnostics
Published on: December 14, 2009
12.0K
Label-Free CD34+ Cell Identification Using Deep Learning and Lens-Free Shadow Imaging Technology
Minyoung Baik1, Sanghoon Shin1, Samir Kumar1
1Department of Electronics and Information Engineering, Korea University, Sejong 30019, Republic of Korea.
Biosensors
|December 22, 2023
Summary
A new deep learning method using lens-free shadow imaging technology (LSIT) accurately identifies CD34+ cells without staining. This portable, user-friendly approach aids leukemia diagnosis and monitoring for non-experts.
Area of Science:
- Biomedical Engineering
- Computational Biology
- Hematology
Background:
- Accurate CD34+ cell quantification is crucial for leukemia diagnosis and monitoring.
- Current methods like flow cytometry are complex, time-consuming, and require specialized resources.
- A need exists for accessible, efficient, and label-free CD34+ cell detection techniques.
Purpose of the Study:
- To develop and validate a novel label-free method for CD34+ cell identification using deep learning and lens-free shadow imaging technology (LSIT).
- To assess the performance of a custom AlexNet model in classifying CD34+ cells compared to existing deep learning architectures and standard methods.
- To evaluate the reliability and correlation of the proposed LSIT-based approach with fluorescence-activated cell sorting (FACS) for CD34+ cell quantification.
Main Methods:
- Bone marrow and peripheral blood samples from leukemia patients were processed to isolate mononuclear cells.
- A proprietary LSIT device (Cellytics) was used to capture images of cells without staining.
- A custom AlexNet deep learning model was trained on the generated dataset to differentiate CD34+ from non-CD34+ cells.
Main Results:
- The custom AlexNet model achieved high accuracy (97.3% training, 96.2% validation) in identifying CD34+ cells from 1929 bone marrow images.
- The AlexNet model demonstrated superior performance compared to Vgg16 and ResNet50 models.
- A strong correlation (R²=0.81) was observed between LSIT-based quantification and FACS, with reliable agreement shown by Bland-Altman analysis.
Conclusions:
- Deep learning-powered LSIT offers a groundbreaking, label-free approach for rapid CD34+ cell classification and quantification.
- This technology enhances accessibility for non-experts, simplifying leukemia diagnostics and monitoring.
- The LSIT method provides a viable, efficient alternative to traditional, complex cell analysis techniques.
Keywords:
CD34+ classificationCellyticsartificial intelligencedeep learninglabel-free cytometrylens-free shadow imagingleukemia diagnosispoint-of-care diagnosis
