Table 2 Endoscopic test set: segmentation and detection performance

From: Deep multimodal state-space fusion of endoscopic-radiomic and clinical data for survival prediction in colorectal cancer

Method

DSC (Seg.)

IoU (Seg.)

Prec. (Det.)

Recall (Det.)

F1 (Det.)

HydraMamba (Ours)

0.856 ± 0.012

0.748 ± 0.015

0.905 ± 0.014

0.931 ± 0.013

0.918 ± 0.012

Segmentation baselines

 DeepLabV3+ (ResNet50)28

0.684 ± 0.028

0.520 ± 0.035

–

–

–

 PraNet (2020)29

0.768 ± 0.024

0.623 ± 0.030

–

–

–

 ColonFormer (2022)30

0.812 ± 0.019

0.684 ± 0.025

–

–

–

 ResPVT (2023)31

0.815 ± 0.020

0.688 ± 0.024

–

–

–

 NA-SegFormer (2024)32

0.821 ± 0.018

0.696 ± 0.022

–

–

–

 PolySegNet (2024)33

0.824 ± 0.019

0.700 ± 0.023

–

–

–

 Polyp–SES (2024)34

0.827 ± 0.017

0.705 ± 0.021

–

–

–

 Hybrid ViT (2024)35

0.822 ± 0.020

0.698 ± 0.024

–

–

–

 Viewpoint-aware (2024)36

0.835 ± 0.016

0.717 ± 0.020

–

–

–

 PraNet-V2 (2025)37

0.831 ± 0.017

0.711 ± 0.022

–

–

–

 ProMamba (2024)38

0.845 ± 0.015

0.731 ± 0.019

–

–

–

 ViM-UNet (2024)39

0.828 ± 0.018

0.709 ± 0.021

–

–

–

Detection baselines

 Faster R-CNN (2015)40

–

–

0.842 ± 0.022

0.848 ± 0.021

0.845 ± 0.020

 ACSNet (2023)41

–

–

0.851 ± 0.020

0.859 ± 0.019

0.855 ± 0.018

 YOLOv8 (2024)42

–

–

0.865 ± 0.018

0.857 ± 0.020

0.861 ± 0.017

 CRH-YOLO (2024)43

–

–

0.872 ± 0.017

0.864 ± 0.019

0.868 ± 0.016

 PolypGen challenge top44

–

–

0.869 ± 0.016

0.871 ± 0.017

0.870 ± 0.015

 YOLOv13 (2025 baseline)45

–

–

0.875 ± 0.015

0.871 ± 0.018

0.873 ± 0.014

Joint seg–det baselines

 MedSAM (2023)46

0.829 ± 0.018

0.708 ± 0.023

0.862 ± 0.019

0.868 ± 0.018

0.865 ± 0.017

 QueryNet (2024)47

0.838 ± 0.017

0.721 ± 0.021

0.884 ± 0.016

0.878 ± 0.017

0.881 ± 0.015

 MedSAM-2 (2024)48

0.841 ± 0.016

0.726 ± 0.020

0.871 ± 0.017

0.879 ± 0.016

0.875 ± 0.014

  1. Segmentation evaluated by Dice similarity coefficient (DSC) and Intersection-over-Union (IoU). Detection evaluated by precision (Prec.), recall (Sens.), and F1-score. All metrics in the table are reported as mean ± 95% confidence interval.
Back to article page