MedSegDiff-V2: Diffusion-Based Medical Image Segmentation with Transformer
Authors: Junde Wu, Wei Ji, Huazhu Fu, Min Xu, Yueming Jin, Yanwu Xu
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We verify its effectiveness on 20 medical image segmentation tasks with different image modalities. Through comprehensive evaluation, our approach demonstrates superiority over prior state-of-the-art (SOTA) methodologies. |
| Researcher Affiliation | Collaboration | Junde Wu1,2,3,4, Wei Ji5, Huazhu Fu6, Min Xu*7,3, Yueming Jin2, Yanwu Xu*8 1University of Oxford 2National University of Singapore 3Mohamed bin Zayed University of Artificial Intelligence 4Kids with Tokens 5University of Alberta 6Institute of High Performance Computing, A*STAR 7Carnegie Mellon University 8Singapore Eye Research Institute |
| Pseudocode | No | The paper describes the methods in prose and equations but does not include structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code is released at https://github.com/Kids With Tokens/Med Seg Diff |
| Open Datasets | Yes | Two datasets are used to verify the general segmentation performance, which are public AMOS2022(Ji et al. 2022) dataset with sixteen anatomies and public BTCV(Fang and Yan 2020) dataset with twelve anatomies annotated for abdominal multi-organ segmentation. The other four public datasets REFUGE2 (Fang et al. 2022), Bra Ts-2021 dataset (Baid et al. 2021), ISIC dataset(Milton 2019) and TNMIX dataset (Pedraza et al. 2015) are used to verify the performance on multi-modal images, which are the optic-cup segmentation from fundus images, the brain tumor segmentation from MRI images, and the thyroid nodule segmentation from ultrasound images. |
| Dataset Splits | No | The paper states that experiments were 'trained/tested' and mentions dataset names, but does not provide specific details on the train, validation, and test dataset splits (e.g., percentages or counts). |
| Hardware Specification | Yes | All experiments were conducted using the Py Torch platform and trained/tested on 4 NVIDIA A100 GPUs. |
| Software Dependencies | No | The paper mentions 'Py Torch platform' and 'Adam W optimizer' but does not specify version numbers for these or other software dependencies. |
| Experiment Setup | Yes | All images were uniformly resized to a resolution of 256 256 pixels. The networks were trained in an end-to-end manner using the Adam W(Loshchilov and Hutter 2017) optimizer with a batch size of 32. The initial learning rate was set to 1 10 4. We employed 100 diffusion steps for the inference. |