Anatomical Knowledge Mining and Matching for Semi-supervised Medical Multi-structure Detection
Authors: Bin Pu, Liwen Wang, Jiewen Yang, Xingbo Dong, Benteng Ma, Zhuangzhuang Chen, Lei Zhao, Shengli Li, Kenli Li
AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments across five publicly available ultrasound datasets demonstrate that Semi-akmm sets a new benchmark in performance with solid results that outperform existing methods. Extensive experiments are performed on five public datasets, and quantitative experimental results and visualization analysis show that our Semi-akmm achieves state-of-the-art performance. |
| Researcher Affiliation | Academia | 1Hunan University, Changsha, China 2Anhui University, Hefei, China 3The Hongkong University of Science and Technology, HKSAR, China 4Shenzhen Maternity and Child Healthcare Hospital, Southern Medical University, Shenzhen, China EMAIL, EMAIL, EMAIL, EMAIL |
| Pseudocode | No | The paper describes methods like ASM and APKT with mathematical formulations (Eq. 1-9) and textual descriptions, but does not include any explicitly labeled pseudocode or algorithm blocks formatted like code. |
| Open Source Code | Yes | Appendices are available at https://github.com/Liwen Wang919/Semiakmm. |
| Open Datasets | Yes | Fetal Cardiac Structure (FCS) (Pu et al. 2024) is a diversified ultrasound dataset collected from our two medical centers... Early Pregnancy View (EPV) (Lin et al. 2022) is a challenging early pregnancy ultrasound dataset collected from different ultrasound devices... |
| Dataset Splits | Yes | The FCS and EPV datasets were divided into train, valid, and test sets in the ratio of 7:1:2, and all the settings remained the same. |
| Hardware Specification | Yes | For a fair comparison, we use Faster-RCNN with FPN (Girshick 2015) and Res Net-50 backbone (He et al. 2016) as the detector, which is implemented in Py Torch and trained 40k iterations with 1 batchsize on one RTX3090 GPU. |
| Software Dependencies | No | The paper mentions 'Py Torch' as the implementation framework but does not specify its version number or versions for any other key software components or libraries. |
| Experiment Setup | Yes | For data augmentation, we use random horizontal flipping for weak augmentation. Based on this augmentation, we then add random color jittering, grayscale, gaussian blurring and cutout patches for strong augmentation... We uniformly resized images to 1024 1024 for all stages, and we trained the model using the stochastic gradient descent (SGD) optimizer with an initial learning rate of 0.01 with the weight decay of 5 10 4. ...we set as 1, 1 and 0.5 in our experiment. ...In all experiments, we set C = 0.2. ...To balance computational efficiency and detection effectiveness, we set K = 20. |