Related Experiment Video
Updated: Jun 23, 2025
09:11
Automation of the Micronucleus Assay Using Imaging Flow Cytometry and Artificial Intelligence
Published on: January 27, 2023
2.1K
An automated in vitro wound healing microscopy image analysis approach utilizing U-net-based deep learning
Dilan Doğru1, Gizem D Özdemir1,2, Mehmet A Özdemir3,4
1Department of Biomedical Engineering, Graduate School of Natural and Applied Sciences, Izmir Katip Celebi University, Izmir, Turkey.
BMC Medical Imaging
|June 24, 2024
Summary
Automated analysis of in vitro wound healing images using U-net deep learning models significantly improves accuracy and speed. These advanced convolutional neural network (CNN) approaches outperform traditional methods for wound area segmentation and therapy efficacy assessment.
Area of Science:
- Computational biology
- Medical imaging analysis
- Machine learning in drug discovery
Background:
- Accurate assessment of in vitro wound healing images is crucial for evaluating therapeutic efficacy.
- Current manual and semi-automated methods are user-dependent, time-consuming, and lack sensitivity.
- Automated analysis approaches are needed to overcome these limitations.
Purpose of the Study:
- To develop and evaluate U-net based convolutional neural network (CNN) models for automated segmentation of in vitro wound healing images.
- To compare the performance of these deep learning models against conventional tools like ImageJ and TScratch.
- To establish a more sensitive and efficient method for calculating wound areas and assessing therapy efficacy.
Main Methods:
- Implemented three U-net architecture variations (U-net, U-net++, Attention U-net) using CNNs for image segmentation.
- Applied a novel augmentation method for enhanced edge analysis during preprocessing.
- Utilized two independent datasets for model training and validation.
- Calculated wound areas from predicted masks and compared results with ImageJ and TScratch.
Main Results:
- U-net based models achieved high segmentation accuracy with average Dice Similarity Coefficient (DSC) scores of 0.958-0.968.
- The developed models demonstrated significantly lower percentage errors (3.70-6.41%) compared to ImageJ (22.59%) and TScratch (33.88%).
- Deep learning models offered substantial improvements in analysis time and segmentation sensitivity.
Conclusions:
- The developed U-net based models significantly outperform conventional methods in terms of speed and accuracy for in vitro wound healing image analysis.
- These models show great potential for reliable in vitro wound area prediction across various experimental conditions.
- The automated approach offers a robust solution for drug discovery and therapeutic evaluation.
