Related Experiment Video
Updated: Mar 28, 2026
08:40
Quantitation of Protein Expression and Co-localization Using Multiplexed Immuno-histochemical Staining and Multispectral Imaging
Published on: April 8, 2016
13.5K
Prostate cancer tissue mapping and stratification using DRAQ5 and Eosin fluorescent labels integrated with AI
Michail Georgios Papachristos1, Emiliano Spezi2, Carolina Fuentes3
1School of Medicine, Division of Cancer and Genetics, Cardiff University, Cardiff, United Kingdom.
Plos One
|March 26, 2026
Summary
This study developed AI models using DRAQ5 and Eosin fluorescent probes for prostate cancer grading and segmentation. The models show promise for clinical diagnostics, demonstrating robustness against imaging variability.
Area of Science:
- Biomedical Imaging
- Artificial Intelligence in Pathology
- Computational Biology
Background:
- Fluorescent microscopy with DRAQ5 and Eosin probes enables rapid, H&E-like pseudoimage generation for tissue characterization.
- These probes offer specific nuclear and cytoplasmic features, enhancing spatial resolution and multi-parametric analysis.
- Understanding the impact of image acquisition variability on AI accuracy is crucial for clinical translation.
Purpose of the Study:
- To develop and evaluate deep learning models for classifying and segmenting prostate cancer tissue labeled with DRAQ5 and Eosin.
- To systematically investigate the influence of image acquisition parameters on AI model performance and robustness.
- To establish the reliability of an automated Gleason grading pipeline for prostate cancer.
Main Methods:
- Prostate tissue samples were labeled using DRAQ5 (DNA probe) and Eosin, generating a two-channel fluorescent readout.
- Deep learning networks were utilized to classify and segment fluorescent image regions into healthy, low-grade, and high-grade cancerous tissue.
- Image acquisition variability (focus, noise, zoom, lens) was induced to assess the robustness and reproducibility of the AI pipeline.
Main Results:
- Machine learning classifiers achieved high Area Under the Curve (AUC) values for cancer grading (e.g., 0.9314 for High Grade vs. Healthy).
- Pixel-wise cancer segmentation demonstrated strong performance, with DICE scores up to 0.8436 for background and 0.705 for cancerous tissue.
- The segmentation model exhibited robustness against various induced image acquisition variations.
Conclusions:
- DRAQ5 and Eosin labeling combined with AI presents a viable pipeline for diagnostic applications in fluorescent imaging.
- The developed methodology shows potential for automated Gleason grading and tissue segmentation in clinical settings.
- Further research is warranted to advance this fluorescent biomarker and AI approach for clinical implementation.
