Enhancing Image Quality and Early Gastric Cancer Detection Using Pre-Processing Techniques on the Kvasir-SEG Dataset

The main reason for a large portion of deaths related to cancer is caused by gastric cancer which is the worst form of disease. Endoscopy is the test that can identify this disease at its early stage which can result in patient survival rates. This study tries to enhance the processing of gastroscopic images by using an enhanced Deep learning (DL) model to boost the accuracy of stomach cancer. Pre-processing frequently involves resizing and bilaterally filtering images to increase clarity and reduce noise. Gamma correction is then used to enhance contrast and make the images simpler to read. To function the model more consistently approaches like zooming, scaling, and rotation are applied to improve the images. The model was trained and evaluated in this research utilizing the Kvasir-SEG dataset, which contains thorough annotations for gastroscopic images. To gauge the outcomes, two important metrics were employed: Mean Squared Error (MSE) and Peak Signal-to-Noise Ratio (PSNR). The processed photos show a very little decrease in quality, with Peak Signal-to-Noise Ratio PSNR values ranging from 50.65 to 52.94. The remarkably low Mean Squared Error (MSE) values (ranging from 0.0023 to 0.0029) show that the treated images are still fairly close to the originals. Based on the available data, this approach could significantly enhance the early detection of stomach cancer, allowing doctors to enhance and expedite patient care. In the future, an improved DL model will be used for lesion segmentation.