EVALUATING THE EFFECTIVENESS OF PRE-TRAINED ENCODERS IN SEMANTIC SEGMENTATION FOR POLYP DETECTION
Rohit Srivastava,Manish Madhava Tripathi
Department of Computer Science and Engineering, Integral University, Lucknow, Uttar Pradesh – 226026, India
Abstract: In this study, we explore the use of pre-trained encoders in semantic segmentation models for polyp detection within medical imaging, specifically focusing on endoscopic images. Polyp segmentation plays a crucial role in early diagnosis and treatment planning for colorectal cancer, making accurate segmentation models essential. Using U-Net as a base architecture, this study investigates the impact of several pre-trained encoders, such as VGG-16, ResNet-50, and VGG-19, on segmentation accuracy. Evaluation metrics, including Dice Coefficient, Intersection over Union (IoU), allow us to assess performance comprehensively. Results show that ResNet-50 achieves superior accuracy in capturing polyp boundaries, while lighter encoders like VGG-16 and VGG-19, excel in speed and are promising for real-time applications.
Keywords: Polyp Detection, Semantic Segmentation, U-Net Architecture