Dual-SAM: Prompt-enhanced Dual-branch SAM for Robust Semi-Supervised Medical ImageSegmentation

Overall Framework

Setup

git clone https://github.com/Heirudy/Dual-SAM.git

Getting Started

1. Install Environment

see requirements.txt

2. Data Preparation

The structure is as follows.

Dual-SAM
├── datasets
│   ├── image
│     ├── image0.png
|     ├── ...
|   ├── mask
│     ├── image0.png
|     ├── ...

3. train

For 3D datasets

python tranin.py 

For 2D datasets

python tranin_2D.py 

Supplementary Materials

Codebase and all experimental logs

The comparative experiment codebase and its log files are available in the "Comparison Method" folder

Additional experimental details

1. Sensitivity analysis of loss weights λ₁ and λ₂

2. Comparative Visualizations of LEPF (Fourier Transform) vs. Wavelet Transform

3. Step-by-step pseudocode for Cross-Path Prompt Fusion (CPF)

4. Visual comparisons of predicted masks from two decoders at multiple training stages

5. Sensitivity analysis of different LoRA ranks on Chest X-ray segmentation performance

6. Comparison of cardiac structure segmentation and clinical diagnostic indicators calculation on ACDC dataset.