Gastrointestinal Bleeding Classification in Capsule Endoscopy Images using Deep Learning

Gastrointestinal bleeding represents a critical health issue, underscoring the pressing need for precise and timely identification of bleeding or lesions within wireless capsule endoscopy (WCE) images to facilitate effective clinical intervention. This paper addresses this imperative by delving into an extensive exploration of various Convolutional Neural Network (CNN) architectures tailored for the classification of gastrointestinal bleeding in WCE images.To comprehensively assess the efficacy of different CNN architectures in handling the nuances of WCE images, a diverse array of models are implemented. These include well-established architectures such as Vanilla CNN, Alexnet, Bottleneck, DenseNet-121, GoogLeNet, ResNet-50, VGG16, VGG19, Attention-based CNN, Capsule Networks (CapsNets). Each architecture undergoes meticulous scrutiny to discern its specific strengths and weaknesses in accurately classifying instances of gastrointestinal bleeding.This paper includes valuable insights and a benchmark for future advancements in the pursuit of accurate and efficient diagnosis through WCE images. Significantly, the outcomes of these assessments are provided, providing an in-depth analysis of how well the applied CNN architectures classified gastrointestinal bleeding. This comprehensive analysis contributes significantly to the understanding of the strengths and capabilities of each model, providing a foundation for further research and optimization in the field.