HRSeg: Leveraging High-Resolution Images to Enhance Polyp Segmentation Quality

The use of Artificial Intelligence (AI) to segment polyps in medical imaging has the ability to significantly enhance the accuracy and speed of diagnoses, ultimately leading to improved patient outcomes. However, current segmentation methods for polyp images have limitations in accurately segmenting small polyps due to lack of data and the complexity of distinguishing polyp tissues from their surrounding environments. To tackle this issue, we proposed a novel approach named HRSeg, which exploits both high resolution (HR) and low resolution (LR) images to provide comprehensive information and context, resulting in improved segmentation accuracy. Our method achieves an average increase of 0.3% in mDice when being trained with the same dataset as the state-of-the-art model. Furthermore, when being trained with augmented data, our HRSeg approach demonstrates an average increase of 3.2% in mDice. Our implementation is available at https://github.com/dranhclub/HRSeg.