Hybrid of DNN Feature Extraction and Ensemble Classification for Identification of Esophagitis and Barretts in Upper Gastrointestinal Tract Images

The main focus of this work is to perform a computer vision classification method for upper gastrointestinal tract image analysis with a pre-trained deep learning features extraction stage. An open-source dataset was used containing images of upper gastrointestinal tract problems including Esophagitis (260 images) and Barretts (94 images). In the given input, image pre-processing with image re-sizing, and noise removal was applied, and then finally working extraction of features using pre-trained deep neural networks. The extracted data from the second last layer of pre-trained models represents the most prominent features. The extracted features are then classified with the help of machines and ensemble learning methods. The improved classification results have been identified using pre-trained models based on feature extraction techniques in comparison to traditional deep learning models.