| Summary: | Today, cancer is one of the leading causes of death in humans in the world. Cancers affect different parts of the human anatomy in different ways. There are significantly more deaths associated with gastrointestinal cancers such as colorectal cancer than with all other cancers. With computer-aided diagnosis systems, medical practitioners can recognize a variety of illnesses more quickly than with manual procedures. It is possible to reduce mortality significantly by detecting and removing precancerous lesions early. Due to the time-consuming and challenging nature of manual diagnosis, researchers have developed computational algorithms that are used to identify and classify diseases such as gastrointestinal (GI) diseases. Computer-aided diagnosis relies heavily on the classification of medical images, which is a challenging task. Therefore, this study presents a less sophisticated but still effective pre-processing technique for identifying endoscopic images known as denoising capsule networks (Dn-CapsNets). Moreover, we constructed activation maps (AM) using the feature representations to visualize the results. As a result of these evaluations, the trained model achieved 94.16%, 83.1%, 86.7%, 96.1%, 86.6%, and +0.69 respectively in terms of accuracy, precision, sensitivity, specificity, F1-score, and Matthew’s correlation. Comparisons between the proposed method and the current state-of-the-art have shown improved accuracy.
|
|---|