ElixirNet: Relation-Aware Network Architecture Adaptation for Medical Lesion Detection
Authors: Chenhan Jiang, Shaoju Wang, Xiaodan Liang, Hang Xu, Nong Xiao11093-11100
AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on Deep Lesion and Kits19 prove the effectiveness of Elixir Net, achieving improvement of both sensitivity and precision over FPN with fewer parameters. |
| Researcher Affiliation | Collaboration | Chenhan Jiang,2 Shaoju Wang,1 Hang Xu,2 Xiaodan Liang,1 Nong Xiao1 1Sun Yat-Sen University, 2Huawei Noah s Ark Lab |
| Pseudocode | No | The paper describes its methods verbally and with architectural diagrams but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide concrete access to source code for the methodology described, nor does it explicitly state that the code is released or available in supplementary materials. |
| Open Datasets | Yes | We conduct experiments on the Deep Lesion (Yan et al. 2018a) and Kits19 (Heller et al. 2019) datasets. |
| Dataset Splits | Yes | We select 30% as validation (4889 lesions) and test (4927 lesions), while the rest is regarded as the training set (22919 images) following the official division (Yan, Bagheri, and Summers 2018; Yan et al. 2018a). |
| Hardware Specification | Yes | We conduct all experiments on a single server with 4 GTX1080 cards in Pytorch (Paszke et al. 2017). |
| Software Dependencies | No | The paper mentions 'Pytorch (Paszke et al. 2017)' but does not provide specific version numbers for Pytorch or other ancillary software components. |
| Experiment Setup | Yes | Network weights ωis updated after training 15 epochs with batch size =16 (same with validation sets). We choose momentum SGD with initial learning rate 0.02 (annealed down to zero following a cosine schedule), momentum 0.9, and weight decay 3 10 4 as an optimizer for weights ω, while architecture weights α are optimized by Adam with initial learning rate 3 10 4 , momentum 0.999 and weight decay 10 3. |