Related Experiment Video
Updated: Oct 22, 2025
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
[Automatic segmentation of kidney tumor based on cascaded multiscale convolutional neural networks]
Hong Ji1, Xusheng Qian2, Zhiyong Zhou2,3
1School of Biological Science and Medical Engineering, Southeast University, Nanjing 210096, P.R.China.
Abstract:
The background of abdominal computed tomography (CT) images is complex, and kidney tumors have different shapes, sizes and unclear edges. Consequently, the segmentation methods applying to the whole CT images are often unable to effectively segment the kidney tumors. To solve these problems, this paper proposes a multi-scale network based on cascaded 3D U-Net and DeepLabV3+ for kidney tumor segmentation, which uses atrous convolution feature pyramid to adaptively control receptive field. Through the fusion of high-level and low-level features, the segmented edges of large tumors and the segmentation accuracies of small tumors are effectively improved. A total of 210 CT data published by Kits2019 were used for five-fold cross validation, and 30 CT volume data collected from Suzhou Science and Technology Town Hospital were independently tested by trained segmentation models. The results of five-fold cross validation experiments showed that the Dice coefficient, sensitivity and precision were 0.796 2 ± 0.274 1, 0.824 5 ± 0.276 3, and 0.805 1 ± 0.284 0, respectively. On the external test set, the Dice coefficient, sensitivity and precision were 0.817 2 ± 0.110 0, 0.829 6 ± 0.150 7, and 0.831 8 ± 0.116 8, respectively. The results show a great improvement in the segmentation accuracy compared with other semantic segmentation methods.
Insights
This study introduces a novel multi-scale network for segmenting kidney tumors in computed tomography (CT) images. The proposed method significantly improves segmentation accuracy for both large and small tumors, addressing challenges with complex backgrounds and unclear tumor edges.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Abdominal computed tomography (CT) imaging presents complex backgrounds, making kidney tumor segmentation challenging due to variations in shape, size, and unclear edges.
- Existing segmentation methods often struggle with the intricacies of CT images, leading to suboptimal kidney tumor segmentation.
Purpose of the Study:
- To develop an advanced multi-scale network for accurate kidney tumor segmentation in abdominal CT images.
- To enhance the segmentation of both large and small kidney tumors by improving edge definition and overall accuracy.
Main Methods:
- Proposed a novel multi-scale network integrating cascaded 3D U-Net and DeepLabV3+ architectures.
- Employed an atrous convolution feature pyramid to adaptively control the receptive field for improved feature extraction.
- Fused high-level and low-level features to enhance segmentation of diverse tumor characteristics.
Main Results:
- Achieved a Dice coefficient of 0.796 ± 0.274, sensitivity of 0.824 ± 0.276, and precision of 0.805 ± 0.284 in five-fold cross-validation.
- On an external test set, the model yielded a Dice coefficient of 0.817 ± 0.110, sensitivity of 0.829 ± 0.150, and precision of 0.831 ± 0.116.
- Demonstrated significant improvements in segmentation accuracy compared to other semantic segmentation methods.
Conclusions:
- The proposed multi-scale network effectively addresses the challenges of kidney tumor segmentation in complex CT images.
- The method shows robust performance in improving segmentation accuracy for tumors of various sizes and complexities.
- This approach offers a promising advancement for automated kidney tumor detection and analysis in clinical settings.
