Squeeze and multi-context attention for polyp segmentation

摘要

Artificial Intelligence-based Computer Aided Diagnostics (AI-CADx) havebeen proposed to help physicians in reducing misdetection of polyps in colonoscopyexamination. The heterogeneity of a polyp's appearance makes detectionchallenging for physicians and AI-CADx. Towards building better AICADx,we propose an attention module called Squeeze and Multi-ContextAttention (SMCA) that re-calibrates a feature map by providing channel andspatial attention by taking into consideration highly activated features and contextof the features at multiple receptive fields simultaneously. We test theeffectiveness of SMCA by incorporating it into the encoder of five popular segmentationmodels. We use five public datasets and construct intra-dataset andinter-dataset test sets to evaluate the generalizing capability of models withSMCA. Our intra-dataset evaluation shows that U-Net with SMCA and withoutSMCA has a precision of 0.86 ± 0.01 and 0.76 ± 0.02 respectively on CVC-ClinicDB.Our inter-dataset evaluation reveals that U-Net with SMCA and withoutSMCA has a precision of 0.62 ± 0.01 and 0.55 ± 0.09 respectively when trainedon Kvasir-SEG and tested on CVC-ColonDB. Similar results are observed usingother segmentation models and other public datasets. In conclusion, we demonstratethat incorporating SMCA into the segmentation models leads to anincrease in generalizing capability of the segmentation models.