Rethinking Semi-Supervised Medical Image Segmentation: A Variance-Reduction Perspective
Authors: Chenyu You, Weicheng Dai, Yifei Min, Fenglin Liu, David Clifton, S. Kevin Zhou, Lawrence Staib, James Duncan
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we experimentally validate our approaches on eight benchmarks, i.e., five 2D/3D medical and three semantic segmentation datasets, with different label settings, and our methods consistently outperform state-of-the-art semi-supervised methods. |
| Researcher Affiliation | Academia | 1Yale University 2University of Oxford 3University of Science and Technology of China |
| Pseudocode | No | The paper describes methods and processes through text and mathematical equations but does not include any clearly labeled 'Pseudocode' or 'Algorithm' blocks. |
| Open Source Code | Yes | 1Codes are available on here. 5Codes are available on here. |
| Open Datasets | Yes | Our experiments are conducted on five 2D/3D representative datasets in semi-supervised medical image segmentation literature, including 2D benchmarks (i.e., ACDC [81], Li TS [82], and MMWHS [83]) and 3D benchmarks (i.e., LA [84] and in-house MP-MRI). |
| Dataset Splits | Yes | The ACDC dataset... We utilize 120, 40, and 40 scans for training, validation, and testing. |
| Hardware Specification | Yes | All experiments are conducted with Py Torch [101] on an NVIDIA RTX 3090 Ti. Hardware: Single NVIDIA Ge Force RTX 3090 GPU; |
| Software Dependencies | Yes | All experiments are conducted with Py Torch [101] on an NVIDIA RTX 3090 Ti. Software: Py Torch 1.10.2+cu113, and Python 3.8.11). |
| Experiment Setup | Yes | We adopt an SGD optimizer with momentum 0.9 and weight decay 10 4. The initial learning rate is set to 0.01. For pre-training, the networks are trained for 100 epochs with a batch size of 6. As for fine-tuning, the networks are trained for 200 epochs with a batch size of 8. The learning rate decays by a factor of 10 every 2500 iterations during the training. We apply the temperature with τt =0.01, τs =0.1, and τ =0.5, respectively. The size of the memory bank is set to 36. For the CL training, we use the implementation from [17] and leave all parameters on their default settings, e.g., we apply the hyperparameters with λ1 =0.01, λ2 =1.0, and λ3 =1.0. |