AE-FLOW: Autoencoders with Normalizing Flows for Medical Images Anomaly Detection
Authors: Yuzhong Zhao, Qiaoqiao Ding, Xiaoqun Zhang
ICLR 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental evaluation on four medical and one non-medical images datasets showed that the proposed model outperformed the other approaches by a large margin, which validated the effectiveness and robustness of the proposed method. |
| Researcher Affiliation | Academia | Yuzhong Zhao, Qiaoqiao Ding, Xiaoqun Zhang School of Mathematical Sciences, MOE-LSC and Institute of Natural Sciences Shanghai Jiao Tong University {zhaoyuzhong, dingqiaoqiao, xqzhang}@sjtu.edu.cn |
| Pseudocode | No | The paper describes the model architecture and loss functions, but it does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper cites a third-party library's GitHub URL (FrEIA) but does not provide a specific link or explicit statement for the open-sourcing of their own AE-FLOW implementation code. |
| Open Datasets | Yes | The proposed method was performed on five datasets with image-level annotations: the OCT (Optical Coherence Tomography) dataset, the Chest X-ray dataset (Kermany et al., 2018), the skin image ISIC2018 ( International Skin Imaging Collaboration) dataset (Tschandl et al., 2018; Codella et al., 2019), the brain tumor Bra TS2021 (Brain Tumor Segmentation) dataset (Menze et al., 2014; Bakas et al., 2017; Baid et al., 2021) and the microscopic images MIIC (Microscopic Images of Integrated Circuits) dataset (Huang et al., 2021). |
| Dataset Splits | Yes | In Appendix A: "We performed five-fold cross-validation experiments for all four medical datasets to further evaluate the robustness and effectiveness of our proposed method. We randomly split the initial test set into validation and test sets in the ratio of 2:8. In order to maintain the balance of each category, we keep the original percentage of normal and abnormal data in each set." |
| Hardware Specification | No | The paper mentions "the Student Innovation Center at Shanghai Jiao Tong University for providing us the computing services" but does not specify any particular hardware (e.g., GPU/CPU models, memory) used for the experiments. |
| Software Dependencies | No | The paper mentions using the "FrEIA library (Ardizzone et al., 2018-2022)" for the flow architecture and the "Adam method (Kingma & Ba, 2014)" as the optimizer, but it does not specify exact version numbers for software dependencies or libraries. |
| Experiment Setup | Yes | Optimizer: The momentum, batch size, learning rate and weight decay hyperparameters were set to be 0.9, 64, 2 10 4 and 10 5 respectively. The last two hyperparameters were set to be 10 3 and 0 for the Chest X-ray dataset. The model was trained for 100 epochs. Weight parameters: For the loss function, we find the model attains the best results when weight parameters α = 0.5 and β = 0.9. |