Related Experiment Video
Updated: Oct 10, 2025
07:53
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
1.6K
Predicting radiation pneumonitis with fuzzy clustering neural network using 4DCT ventilation image based dosimetric
Peng Huang1, Hui Yan1, Zhihui Hu1
1Department of Radiation Oncology, National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Beijing, China.
Quantitative Imaging in Medicine and Surgery
|December 10, 2021
Summary
Predicting radiation-induced pneumonitis (RP) in thoracic cancer patients is improved using ventilation image (VI)-based dosimetric parameters. A fuzzy clustering neural network enhances prediction accuracy for lung toxicity risk after radiotherapy.
Area of Science:
- Medical Physics
- Radiation Oncology
- Computational Biology
Background:
- Radiation-induced pneumonitis (RP) is a significant toxicity in thoracic cancer radiotherapy.
- Accurate prediction of RP is crucial for optimizing treatment plans and patient outcomes.
Purpose of the Study:
- To develop a fuzzy clustering neural network model for predicting RP.
- To utilize four-dimensional computed tomography (4DCT) ventilation image (VI)-based dosimetric parameters for prediction.
- To compare the efficacy of VI-based versus structure-based dosimetric parameters.
Main Methods:
- Retrospective calculation of VI from pre-treatment 4DCT data using deformable image registration (DIR) and an improved VI algorithm.
- Derivation of dose-function histogram (DFH) and calculation of dose-function metrics from VI.
- Principal component analysis (PCA) for feature dimension reduction and fuzzy c-means (FCM) for clustering.
- Training a neural network to correlate dosimetric parameters with RP occurrence.
Main Results:
- PCA identified 5 principal components, explaining over 98% of the data variance.
- The optimal number of clusters determined by validity indexes was 4.
- The VI-based model achieved an area under the curve (AUC) of 0.77, outperforming the structure-based model's AUC of 0.67.
Conclusions:
- VI-based dosimetric features are more relevant to lung function and offer higher RP prediction accuracy than structure-based features.
- Fuzzy clustering neural network improves RP prediction accuracy compared to conventional neural networks.
- Combining VI-based dose-function metrics with a fuzzy clustering neural network provides an effective model for assessing lung toxicity risk.
