Advancing Barrett's Esophagus Segmentation: A Deep-Learning Ensemble Approach with Data Augmentation and Model...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Barrett's esophagus requires accurate segmentation for diagnosis and treatment.
• Deep learning models, particularly U-Net variants, show promise for medical image segmentation.

Purpose of the Study:

• To investigate and optimize deep learning-based segmentation of Barrett's esophagus.
• To evaluate the impact of different U-Net backbone architectures and ensemble strategies on segmentation performance.

Main Methods:

• Exploration of U-Net models with various backbone architectures (e.g., DenseNet).
• Application of rigorous data augmentation and ensemble techniques.
• Fine-tuning ensemble weights using grid search and comparison with Deeplabv3+.

Main Results:

• DenseNet backbones demonstrated superior performance in segmentation accuracy.
• Tailored data augmentation and training U-Net models from scratch proved effective.
• Ensemble methods consistently enhanced segmentation, achieving a mean intersection over union (IoU) score of approximately 0.94.

Conclusions:

• Deep learning, especially U-Net with DenseNet backbones and ensemble strategies, offers a robust approach for Barrett's esophagus segmentation.
• Optimized data augmentation and ensemble techniques are crucial for achieving high segmentation accuracy.
• This research advances automated analysis in gastrointestinal diagnostics.