Deep Implicit Statistical Shape Models for 3D Medical Image Delineation
Authors: Ashwin Raju, Shun Miao, Dakai Jin, Le Lu, Junzhou Huang, Adam P. Harrison2135-2143
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Intra-dataset experiments on the task of pathological liver segmentation demonstrate that DISSMs can perform more robustly than four leading FCN models, including nn U-Net + an adversarial prior: reducing the mean Hausdorff distance (HD) by 7.5-14.3 mm and improving the worst case Dice-Sørensen coefficient (DSC) by 1.2-2.3%. More critically, cross-dataset experiments on an external and highly challenging clinical dataset demonstrate that DISSMs improve the mean DSC and HD by 2.1-5.9% and 9.9-24.5 mm, respectively, and the worst-case DSC by 5.4-7.3%. Supplemental validation on a highly challenging and low-contrast larynx dataset further demonstrate DISSM s improvements. |
| Researcher Affiliation | Collaboration | Ashwin Raju 1,2, Shun Miao 1, Dakai Jin 1, Le Lu 1, Junzhou Huang 2, Adam P. Harrison 1* 1 PAII Inc, Bethesda, MD, USA 2 University of Texas at Arlington, Arlington, TX, USA |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Our code is publicly shared1. 1https://github.com/Ash Stuff/dissm |
| Open Datasets | Yes | Liver Dataset: We focus on delineating pathological livers from venous-phase CTs. We use the size-131 training set of the MSD liver dataset (Simpson et al. 2019)... To do this, we also evaluate on the external test set of Raju et al. (2020a)... Larynx Dataset: We also perform validation on larynx segmentation from CT... SOARS method and dataset of Guo et al. (2020) (142 CTs) |
| Dataset Splits | Yes | We use the size-131 training set of the MSD liver dataset (Simpson et al. 2019), splitting it randomly into training, testing, and validation using proportions of 70%, 20%, and 10%, respectively. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications. It only mentions the type of models used (e.g., '3D U-Net'). |
| Software Dependencies | No | The paper mentions general software concepts like 'modern deep learning software' and specific model architectures (e.g., 'U-Net', 'FCNs'), but it does not specify software dependencies with version numbers (e.g., Python 3.x, PyTorch 1.x, CUDA 11.x). |
| Experiment Setup | Yes | The shape decoder structure and hyperparameters follow that of Park et al. (2019) and we use a size 256 latent variable. For the pose encoders, gθE(., .), use the 3D encoder of a 3D U-Net (C ic ek et al. 2016), with 4 downsampling layers and global averaging pooling to produce τ and λ. ... The number of inverted episodic steps, T, for training the translation, scale, rotation, and non-rigid encoders was 7, 15, 15, and 15, respectively. |