Machine learning-based predictive model for stroke lesion characteristics using multimodal MRI(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2026年第2期
- Page:
- 276-280
- Research Field:
- 医学人工智能
- Publishing date:
Info
- Title:
- Machine learning-based predictive model for stroke lesion characteristics using multimodal MRI
- Author(s):
- CHEN Aide; HU Yaheng; WEI Wencun
- Department of Equipment, Ankang Hospital of Traditional Chinese Medicine, Ankang 725000, China
- Keywords:
- Keywords: machine learning multimodal magnetic resonance imaging stroke lesion radiomics
- PACS:
- R318；R743
- DOI:
- DOI:10.3969/j.issn.1005-202X.2026.02.019
- Abstract:
- Abstract: Objective To develop a predictive model for stroke lesion characteristics by integrating multimodal magnetic resonance imaging (MRI) data, radiomics features, and machine learning techniques, and to compare the performance of different machine learning algorithms, including support vector machine, decision tree, extreme gradient boosting (XGBoost), and back propagation neural network, in identifying stroke lesion characteristics, thereby supporting clinical diagnosis and treatment decision-making. Methods Multimodal MRI data were collected from 129 patients admitted to the Neurology Department of a Grade A tertiary hospital between February 2020 and May 2024, including T1-weighted imaging, T2-weighted imaging, diffusion-weighted imaging, and contrast-enhanced imaging. Regions of interest were delineated for each lesion, followed by radiomics feature extraction, with an attention mechanism introduced to enhance feature extraction accuracy. Subsequently, feature selection was conducted using LASSO regression and principal component analysis, and finally, predictive models based on support vector machine, decision tree, XGBoost, and back propagation neural network were constructed using the selected features. Results Model performance was evaluated on an independent test set. Among these models, the LASSO-XGBoost combination outperformed all other combinations across all metrics, whereas the integration of principal component analysis and decision tree had the poorest performance. Feature selection significantly improved the predictive capability of all models, especially for high-dimensional radiomics data processing. Conclusion A predictive model for stroke lesion risk is established by integrating multimodal MRI with machine learning. The LASSO-XGBoost combination exhibits excellent performance in identifying stroke lesion characteristics from multimodal imaging data, demonstrating its promising potential for clinical application.
Last Update: 2026-01-27