<![CDATA[Health involves the proper function of the body and organs, with colon polyps being a common issue. Doctors often face challenges in segmenting medical images, especially endoscopic images for polyp detection. The complexity and variation in the appearance of polyps make accurate identification challenging, and the subjective manual segmentation process can result in misdiagnosis or delayed treatment. This study examines the effectiveness of the combination of U-Net decoder model architecture and VGG19 encoder in segmentation of colon polyp images. This study uses a public dataset, namely Kvasir-Seg with a total of 1000 images of colon polyps. An innovative approach using VGG19 as encoder and U-Net as decoder improves colorectal polyp segmentation, achieving high performance with a Loss of 0.05, Accuracy 0.95, Precision 0.96, Recall 0.92, IoU 0.89, and Dice 0.94. Using optimal parameters such as Nadam Optimizer, 5 Fold Cross Validation, Learning Rate 0.0001, and 25 Epochs significantly improved performance, increasing the Dice Coefficient to 0.92 and IoU to 0.86 compared to previous studies. This study concludes that the proposed architecture is reliable for colon polyp segmentation. Future work should explore attention mechanisms or transformer-based models to enhance accuracy and efficiency.]]>
Copyrights © 2026
