Barrett's Esophagus Segmentation using U-Net Model based on Transfer Learning

In this paper, we propose various deep learning-based U-Net architecture with transfer learning for Barrett's Esophagus segmentation. Various pre-trained models, including VGG19, DenseNet169 and DenseNet201 are employed in the study. We also utilized data preprocessing techniques, such as cropping, GaussianBlur, Gaussian noise, and data augmentation, to increase the amount of available training data. To evaluate the performance of the proposed models, a dataset of Barrett's Esophagus images are adopted and the results show that four best models achieved a meanIoU score of around 0.8, which is a significant improvement over other models. Our findings suggest that using U-Net with transfer learning as a backbone, along with appropriate data pre-processing techniques, can be an effective approach for Barrett's Esophagus segmentation. This paper can serve as a valuable resource for medical professionals seeking to implement accurate and efficient segmentation algorithms for Barrett's Esophagus.