Automated Diagnosis of Lung and Colorectal Pathologies Using a Shallow Capsule Network Model

One of the most hazardous diseases is cancer caused by many biochemical abnormalities. Among all the cancers, lung and colon cancer are the most common and tragic diseases. Effective detection of thoracic and colorectal pathology is vital for timely diagnosis and treatment. In this study, we propose a shallow capsule network model for detecting malignancies in lung and colorectal imaging. Our proposed model is trained and tested on three different datasets: LC25000, IQ-OTHNCCD Lung Cancer Dataset and Chest CT-Scan Dataset with varying train and test data ratios. Despite its shallow architecture, the proposed model achieves high accuracy, with test accuracy metrics of 99.32%, 89.05%, and 99.09% on the respective datasets. We have also shown that the proposed capsule network outperforms traditional deep learning models using less training data. Our findings show that the suggested shallow capsule network model is effective in identifying lung and colorectal disease.