Skip to content

Latest commit

 

History

History
57 lines (42 loc) · 3.53 KB

README.md

File metadata and controls

57 lines (42 loc) · 3.53 KB

● SPA: Efficient User-Preference Alignment against Uncertainty in Medical Image Segmentation

Discord License

SPA, is an advanced segmentation framework that efficiently adapts to diverse test-time preferences with minimal human interaction. By presenting users a select few, distinct segmentation candidates that best capture uncertainties, it reduces clinician workload in reaching the preferred segmentation. This method is elaborated on the paper SPA: Efficient User-Preference Alignment against Uncertainty in Medical Image Segmentation and SPA webpage.

🔥 A Quick Overview

Our uncertainty-aware interactive segmentation model, SPA, efficiently achieves preference-aligned segmentation by incorporating medical image uncertainties and human interactions. Clinicians are presented with one recommended prediction and a few distinct segmentation candidates that capture uncertainty, allowing them to select the one best aligned with their clinical needs. If the user is unsatisfied with the recommended prediction, the model learns from the user selection, adapts itself, and presents users a new set of candidates. Our approach minimizes user interactions and eliminates the need for painstaking pixel-wise adjustments compared to conventional interactive segmentation models.

🧐 Requirement

Install the environment:

conda env create -f environment.yml

conda activate SPA

Further Note: We tested on the following system environment and you may have to handle some issue due to system difference.

Operating System: Ubuntu 22.04
Conda Version: 23.7.4
Python Version: 3.12.4

🎯 Example Cases

Step1: Download MedSAM or SAM pretrained weight and put in the ./checkpoint/sam/ folder, create the folder if it does not exist ⚒️ MedSAM checkpoint

SAM checkpoint

Step2: Download REFUGE2 (update later) or your own multi-rater dataset and put in the data folder, create the folder if it does not exist ⚒️

Step3: Run the training by: python train.py -net sam -mod sam -exp_name 'REFUGE_SPA' -sam_ckpt ./checkpoint/sam/medsam_vit_b.pth -image_size 512 -out_size 256 -b 4 -val_freq 1 -dataset REFUGE -data_path './data/REFUGE'

Step4: Run the validation by: python val.py -net sam -mod sam -exp_name 'val' -vis 1 -sam_ckpt CHECKPOINT_PATH -weights CHECKPOINT_PATH -image_size 512 -out_size 256 -b 1 -val_freq 1 -dataset REFUGE -data_path './data/REFUGE'

🚨 News

  • 25-01-06. Code Uploaded 👩‍💻
  • 24-12-02. SPA's website was released 🤩

📝 Cite

@misc{zhu_spa_2024,
     title={SPA: Efficient User-Preference Alignment against Uncertainty in Medical Image Segmentation},
     author={Zhu, Jiayuan and Wu, Junde and Ouyang, Cheng and Kamnitsas, Konstantinos and Noble, Alison},
     url = {http://arxiv.org/abs/2411.15513},
     doi = {10.48550/arXiv.2411.15513},
     year = {2024},
   }