Gastrointestinal Image Classification Using Deep Learning Architectures Via Transfer Learning

Abstract

Computer aided detection of diseases using machine learning mechanisms on medical images has been an interesting applied research topic in both academia and health sector. Practical studies with the aim of improving the process of decision on the diagnosis of the diseases via accurate classification of the medical images would be benefit of the medical doctors. This paper presents an investigation on the classification of gastrointestinal images using deep learning models. The labeled medical images used in the experiments are publicly available within the Kvasir dataset on Kaggle. The deep learning approaches applied through the experiments are based on the following Convolutional Neural Network architectures used with transfer learning: VGG19, ResNet50V2, ResNet152V2, EfficientNetV2B0, EfficientNetV2B3, InceptionV3, DenseNet201, Xception. The performances of these different architectures on learning the training dataset and classifying the test images are evaluated in terms of the following metrics: accuracy, precision, recall, and F1-score. Regarding the results of the experiments conducted using the same dataset on different deep learning models, VGG19 model outperformed the others with the prediction accuracy ratio of 88.6%.