Automatic classification of upper gastrointestinal tract diseases from endoscopic images

Artificial Intelligence (AI) has played an increasingly crucial part in our daily lives in recent years. Convolutional neural network (CNN) in medical image processing has lately received a lot of interest. With the introduction of modern endoscopic technologies, the doctor could be able to diagnose a patient more accurately. Consequently, it becomes crucial and advantageous to use computer-aided support during procedures. This paper proposes a framework for automatic classification of Upper Gastrointestinal tract diseases that consists of two main parts: a Convolutional Neural Network model (ResNet-50) with Focal Loss application and Data Augmentation techniques that consists of Geometric Transformation (GeoT), Brightness and Contrast Transformation (BaC), which learn hidden features of various Upper GI diseases and anatomical landmark classes, as well as solving the imbalanced dataset problem. As a result, the classification result is improved, with 98.61% of Accuracy on a self-collected dataset from Hanoi Medical University Hospital and Institute of Gastroenterology and Hepatology. Additionally, the proposed method is capable of using in real-time applications by classifying every frame in the video streams.