Deep Learning Approaches to Gastrointestinal Image Classification: Evaluating VGG19 and EfficientNetB0 Models

This research reports on the comparison of a VGG19 and an EfficientNetB0 models for the identification of gastrointestinal images. At the same time that deep learning is becoming the cornerstone of medical imaging applications and as such, our research will aim to select the most accurate, fast, and robust model out of them. By incorporating a Kaggle dataset which include the whole array of gastrointestinal disorders, we resorted to the processes of data augmentation, preparation, and model experimental work involving the training of the model and its testing. It was seen that the functionalities of EfficientNetB0(Test Accuracy-84.4% and Test Loss-0.554)- were much better (in respect to accuracy in classification and computational efficiency) than those of VGG19(Test Accuracy-83.5% and Test Loss-0.905). It is a clear proof of the high generalizability of EfficientNetB0 model from different datasets, probably due to an optimal arrangement of three factors (depth, width and resolution). This work concludes with empathizing the possible use of, EfficientNetB0 in the healthcare space when it comes to diagnosing gastrointestinal problems which therefore results in better patient outcomes. Along with this, future work will look into convergence of these models into clinical practice and envisioning them for medical image classification through the integration of deep learning-based technology.