Related Experiment Video
Updated: Aug 22, 2025
08:58
Label-Free Identification of Lymphocyte Subtypes Using Three-Dimensional Quantitative Phase Imaging and Machine Learning
Published on: November 19, 2018
12.6K
Low-latency label-free image-activated cell sorting using fast deep learning and AI inferencing.
Rui Tang1, Lin Xia2, Bien Gutierrez2
1Department of Electrical and Computer Engineering, University of California, San Diego, La Jolla, CA, 92093, USA; NanoCellect Biomedical Inc., San Diego, CA, 92121, USA.
Biosensors & Bioelectronics
|November 11, 2022
Summary
This study introduces a fast, label-free cell sorting system using deep learning for image-activated cell sorting (IACS). The novel system achieves record-low latency and rapid training, making advanced cell analysis more accessible for biomedical research.
Area of Science:
- Biomedical engineering
- Computational biology
- Cell biology
Background:
- Image-activated cell sorting (IACS) offers insights into biomedical sciences.
- Deep learning (DL) enhances IACS by classifying cells based on complex morphological features.
- Current DL-assisted IACS systems have limitations in capability and implementation, hindering widespread adoption.
Purpose of the Study:
- To develop a novel image-activated cell sorting (IACS) system utilizing fast deep learning algorithms.
- To achieve label-free cell sorting with significantly reduced latency and training times.
- To provide a compact, low-cost, and efficient cell sorting solution for diverse biomedical applications.
Main Methods:
- Implementation of fast deep learning algorithms for cell classification and sorting within an IACS framework.
- Development of an optimized custom Convolutional Neural Network (CNN) UNet architecture.
- Utilizing a simple hardware setup comprising an FPGA, PC, and GPU for efficient computation.
Main Results:
- Achieved an overall sorting latency of less than 3 milliseconds, including signal processing and AI inference.
- Demonstrated rapid deep learning model training in under 30 minutes for a dataset of 20,000 images.
- Exhibited high sorting purity: 96.6% for polystyrene beads and 89.05% (monocytes), 92.00% (lymphocytes), and 98.24% (granulocytes) for human leukocytes.
Conclusions:
- The developed system represents a significant advancement in IACS technology, offering record-breaking speed for AI-driven cell sorting.
- The system's efficiency, low-cost hardware, and label-free operation make it a practical and accessible tool for various research fields.
- This innovative approach overcomes previous limitations, paving the way for broader application of DL-assisted cell sorting in biomedical research.
