מושגי ליבה
Proposing a segmentation model based on Prompt-Mamba for accurate polyp segmentation in medical images.
תקציר
Introduction to the importance of polyp segmentation in cancer prevention.
Challenges in polyp segmentation due to variations in size, shape, and color.
Evolution of deep learning models like U-Net, U-Net++, ResUNet, PraNet, TGANet, and SAM in polyp segmentation.
Introduction of Vision-Mamba technology and prompt-based segmentation.
Detailed explanation of the ProMamba model architecture with Image-encoder, Prompt-encoder, and Mask-decoder.
Utilization of Focalloss and Diceloss in the loss function for improved segmentation quality.
Experimental design, dataset partitioning, baseline comparisons with other methods, and ablation studies.
Results showing superior performance of ProMamba on unseen datasets compared to previous methods.
Conclusion highlighting the significance of incorporating Vision-Mamba and prompt technologies for enhanced generalization capabilities.
סטטיסטיקה
Our smallest model with only 11M parameters is only one-third of the parameters of ParNet and one-fourth of nnU-Net.
ציטוטים
"We are the first to apply Vision-Mamba (ViM) technology to polyp segmentation."
"Our results are 5% better than the performance of the previous best methods on unseen datasets."