Related Experiment Video
Updated: Nov 1, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.1K
DFP-ResUNet:Convolutional Neural Network with a Dilated Convolutional Feature Pyramid for Multimodal Brain Tumor
Jingjing Wang1, Jun Gao1, Jinwen Ren1
1School of Physics and Electronics, Shandong Normal University, Jinan, China.
Computer Methods and Programs in Biomedicine
|June 26, 2021
Summary
This study introduces DFP-ResUNet, an automated neural network for brain tumor segmentation. The model accurately segments enhancing tumor, whole tumor, and tumor core subregions, offering potential for clinical application.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in oncology
- Neurosurgery and oncology
Background:
- Manual brain tumor segmentation is time-consuming and subjective.
- Accurate segmentation of brain tumor subregions is crucial for patient treatment.
- There is a need for automated, reliable methods for brain tumor segmentation.
Purpose of the Study:
- To propose a novel neural network, DFP-ResUNet, for automated segmentation of brain tumor subregions.
- To accurately segment the enhancing tumor (ET), whole tumor (WT), and tumor core (TC).
- To improve the efficiency and objectivity of brain tumor segmentation in clinical practice.
Main Methods:
- Developed a U-Net based neural network incorporating a spatial dilated feature pyramid (DFP) module and residual modules.
- Employed dilated convolutions within the DFP module to extract multiscale image features effectively.
- Utilized a multiclass Dice loss function to address class imbalance issues in segmentation.
- Validated the approach using the Multimodal Brain Tumor Segmentation (BraTS) challenge dataset.
Main Results:
- Achieved high mean Dice scores for ET (0.8431), WT (0.897), and TC (0.9068) on the BraTS 2018 validation set.
- Obtained competitive results on the BraTS 2019 challenge with Dice scores of 0.7985 (ET), 0.90281 (WT), and 0.8453 (TC).
- Demonstrated high sensitivity, specificity, and low Hausdorff distance, indicating robust segmentation performance.
Conclusions:
- The proposed DFP-ResUNet demonstrates significant potential for automated brain tumor subregion segmentation.
- The method shows promise for integration into clinical workflows, improving diagnostic accuracy and treatment planning.
- Ablation experiments confirmed the superiority and feasibility of the developed approach.
