An Accurate Deep Learning-Based Computer-Aided Diagnosis System for Gastrointestinal Disease Detection Using Wireless Capsule Endoscopy Image Analysis

Peptic ulcers and stomach cancer are common conditions that impact the gastrointestinal
(GI) system. Wireless capsule endoscopy (WCE) has emerged as a widely used, noninvasive technique
for diagnosing these issues, providing valuable insights through the detailed imaging of the GI tract.
Therefore, an early and accurate diagnosis of GI diseases is crucial for effective treatment. This paper
introduces the Intelligent Learning Rate Controller (ILRC) mechanism that optimizes the training of
deep learning (DL) models by adaptively adjusting the learning rate (LR) based on training progress.
This helps improve convergence speed and reduce the risk of overftting. The ILRC was applied to
four DL models: EffcientNet-B0, ResNet101v2, InceptionV3, and InceptionResNetV2. These models
were further enhanced using transfer learning, freezing layers, fne-tuning techniques, residual
learning, and modern regularization methods. The models were evaluated on two datasets, the
Kvasir-Capsule and KVASIR v2 datasets, which contain WCE images. The results demonstrated
that the models, particularly when using ILRC, outperformed existing state-of-the-art methods in
accuracy. On the Kvasir-Capsule dataset, the models achieved accuracies of up to 99.906%, and on
the Kvasir-v2 dataset, they achieved up to 98.062%. This combination of techniques offers a robust
solution for automating the detection of GI abnormalities in WCE images, signifcantly enhancing
diagnostic effciency and accuracy in clinical settings.