A Deep Learning Approach to Classify Colon Diseases

Abstract

Early and proper diagnosis of Colon diseases is crucial, as it is essential to successful clinical treatment and patient outcomes. InceptionResNet V2, Traditional convolutional neural network architectures (Xception and ConvNeXt-Tiny) can theoretically be used to label medical imaging. However, they are expensive, uninterpretable, or fail to perform fine-grained discriminative image recognition tasks on complex endoscopic images. We utilize CareNet, a simple yet highly discriminative deep learning model, to address these challenges specifically in colon disease classification. CareNet applies both EfficientNet-B0 (as a baseline, founded on average pooling worldwide, max pooling worldwide, channel-gating and refining convolutional pooling, and an attention-pooling mechanism. This design is more computationally efficient, experienceable in the context of global features, and sensitive to local features. This analysis of the colon endoscopy dataset on a benchmark has identified that CareNet performs well, achieving 99.22% classification accuracy on the colon endoscopy dataset, compared to state-of-the-art models on the same dataset. Moreover, the results of cross-validation prove its strength and ability to generalize to other folds of data. The CareNet, which proposes a solution for clinical decision-support in diagnosing colon disease balance using real-world data, is more precise, effective, and comprehensible compared to existing studies.