Official Pytorch implementation of ConvUNET, from the following paper:
ConvUNET: a Novel Depthwise Separable ConvNet for Lung Nodule Segmentation.
BIBM 2023 (Accepted at 13 Oct. 2023, Regular Paper)
Xinkai Tang, Feng Liu, Ruoshan Kong, Fei Luo, et al.
School of Computer Science, Wuhan University, China
We propose a lightweight depthwise separable convolutional network named ConvUNET, which consists of a Transformer-like encoder and a ConvNet-based decoder. Compared with some Transformer-based models (e.g., SwinUNETR) and ConvNeXt-based models (e.g., 3D UX-Net), our model has the advantages of fewer parameters, faster inference speed, and higher accuracy.
| Models | Params(M) | GFLOPs | DSC(%) | SEN(%) | PPV(%) |
|---|---|---|---|---|---|
| UNETR (Hatamizadeh, 2021) | 92.8 | 82.6 | 86.32 | 87.41 | 88.08 |
| SwinUNETR (Hatamizadeh, 2022) | 62.2 | 328.4 | 88.41 | 89.67 | 89.14 |
| UNETR++ (Shaker, 2023) | 53.5 | 48.0 | 88.42 | 89.57 | 89.08 |
| ConvNeXt V1 (Liu, 2022) | 31.9 | 89.3 | 87.46 | 88.12 | 89.17 |
| ConvNeXt V2 (Woo, 2023) | 31.9 | 89.3 | 87.59 | 88.38 | 88.92 |
| 3D UX-Net (Lee, 2023) | 53.0 | 639.4 | 88.55 | 90.10 | 88.93 |
| ConvUNET (Ours, 2023) | 32.6 | 265.1 | 88.90 | 90.13 | 89.49 |
| Models | Params(M) | GFLOPs | DSC(%) | SEN(%) | PPV(%) |
|---|---|---|---|---|---|
| UNETR (Hatamizadeh, 2021) | 92.8 | 82.6 | 80.42 | 82.04 | 85.13 |
| SwinUNETR (Hatamizadeh, 2022) | 62.2 | 328.4 | 83.65 | 85.22 | 86.48 |
| UNETR++ (Shaker, 2023) | 53.5 | 48.0 | 83.34 | 84.33 | 85.49 |
| ConvNeXt V1 (Liu, 2022) | 31.9 | 89.3 | 81.05 | 82.37 | 85.42 |
| ConvNeXt V2 (Woo, 2023) | 31.9 | 89.3 | 81.10 | 82.62 | 85.05 |
| 3D UX-Net (Lee, 2023) | 53.0 | 639.4 | 83.71 | 85.16 | 86.66 |
| ConvUNET (Ours, 2023) | 32.6 | 265.1 | 84.16 | 85.15 | 87.29 |
This work is supported by National Natural Science Foundation of China (NSFC No.62172309).
If you find this repository helpful, please consider citing:
@inproceedings{tang2023convunet,
title={ConvUNET: a Novel Depthwise Separable ConvNet for Lung Nodule Segmentation},
author={Tang, Xinkai and Liu, Feng and Kong, Ruoshan and Luo, Fei and Huang, Wencai and Zou, Jiani},
booktitle={2023 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)},
pages={1443--1450},
year={2023},
organization={IEEE}
}