GamaComet

GamaComet detects and classifies DNA damage in buccal mucosa cells from comet assay images to provide quantitative assessment of genotoxicity.

Key Features:

• Deep learning framework: Implements Faster R-CNN (Region-based Convolutional Neural Network) for automated detection and classification of comets in comet assay images.
• Automated detection and classification: Performs end-to-end identification and class assignment of comet objects to quantify DNA damage levels.
• Training dataset: Model development used buccal mucosa samples from 24 patients in Indonesia comprising 275 comet assay images containing 519 comets.
• Training enhancements: Employs transfer learning and data augmentation during model training to mitigate limited dataset size.
• Validation and performance: Validated on an independent set from seven patients (43 images, 73 comets) achieving 81.34% detection accuracy and 66.67% classification accuracy.
• Comparative performance: Reported to outperform OpenComet in detection accuracy (OpenComet: 11.5% detection accuracy in the reported comparison).

Scientific Applications:

• DNA damage assessment: Enables quantitative identification and classification of DNA damage in buccal mucosa cells using comet assay imagery.
• Environmental and lifestyle risk analysis: Facilitates investigation of correlations between comet classes and exposures such as smoking for epidemiological studies of genotoxic risk.

Methodology:

Faster R-CNN was trained on a dataset of 275 images (519 comets) from 24 Indonesian patients using transfer learning and data augmentation, and was validated on 43 images (73 comets) from seven patients with performance reported as detection and classification accuracies and compared against OpenComet.