DE-ColonSegNet: An Improved version of ColonSegNet with dual encoder for accurate real-time Polyp segmentation in colonoscopy

High-performance real-time automatic polyp detection is crucial to gastrointestinal healthcare. Efficient detection systems can aid in early diagnosis and intervention, significantly impacting patient outcomes. This study aims to develop an algorithm that strikes a superior balance between computational performance and accuracy compared to existing solutions. Achieving this balance is essential for practical, real-time implementation in clinical settings. The DE-ColonSegNet model architecture is presented as the approach to address the computational performance/accuracy trade-off. This model has been designed to optimize detection accuracy and efficiency, utilizing advanced techniques in deep learning. The DE-ColonSegNet model demonstrates promising outcomes, with an average precision of 0.8845, a mean Intersection over Union (IoU) of 0.7637, and a dice coefficient of 0.8470. These results showcase the model's effectiveness in accurately identifying and segmenting polyps. The robust performance metrics suggest its potential for practical application in real-time clinical settings, contributing to enhanced efficiency in gastrointestinal healthcare.