A Comparative Study of ResNet and DenseNet in the Diagnosis of Colitis Severity

The current diagnostic approaches for assessing the severity of colitis necessitate medical professionals or specialists to subjectively evaluate colitis colonoscopy images, relying extensively on their clinical expertise. The accuracy of these assessments is of utmost importance in guiding subsequent treatment strategies for individuals with colitis. Several deep learning models have demonstrated their efficacy in the domain of medical imaging, serving as dependable tools for visualizing and analyzing medical data. These models include deep learning-based models and convolution-based neural network models. This study aimed to assess the effectiveness of various convolution-based neural network models in diagnosing the severity of colitis. Specifically, the representative ResNet and DenseNet models were chosen for a comparative analysis. Four types of medical imaging images of colitis with different severity were selected for classification and diagnosis. The experimental results demonstrate that DenseNet outperforms ResNet in terms of efficiency and accuracy for diagnosing colitis severity. DenseNet achieves an accuracy rate of up to 80%, indicating the promising potential for its application in the field of medicine.