The　precise　and　reliable　segmentation　of　endoscopic　images　plays　a　pivotal　role　in　the　early　diagnosis　of　gastrointestinal　tract　tumors.　However,　in　comparison　to　standard　RGB　images,　endoscopic　images　exhibit　weaker　contrast　and　more　indistinct　lesion　boundaries,　posing　a　significant　challenge　to　the　accurate　segmentation　of　lesion　regions.　This　paper　introduces　a　novel　and　efficacious　framework,　named　DG-Net,　for　segmenting　lesion　regions　in　endoscopic　images.　DG-Net　is　a　dual-guided　network　comprising　the　bilateral　attention　branch　and　the　boundary　aggregation　branch.　Firstly,　a　mask　decoder　named　Progressive　Partial　Decoder　(PPD)　and　a　module　known　as　Full-context　Bilateral　Relation　(FBR)　are　developed　to　constitute　the　bilateral　attention　branch.　The　primary　objective　of　this　branch　is　to　focus　attention　on　the　ambiguous　boundaries　of　lesion　regions　by　augmenting　the　correlation　between　foreground　and　background　cues　in　the　images.　Subsequently,　a　boundary　decoder　named　Boundary-Aware　Extraction　(BAE)　and　a　module　termed　Boundary-guided　Feature　Aggregation　(BFA)　are　specifically　designed　to　form　the　boundary　aggregation　branch.　This　branch　utilizes　additional　boundary　semantic　cues　to　generate　features　that　accentuate　the　structural　aspects　of　lesion　regions.　Comprehensive　experiments　were　conducted　on　four　endoscopic　image　datasets,　namely　Kvasir-SEG,　ES-Gastric,　CVC-ClinicDB,　and　CVC-ColonDB.　The　results　of　qualitative　and　quantitative　analyses　confirm　the　effectiveness　and　practicality　of　DG-Net.　©　2024　The　Author(s)