研究目的
To predict the molecular subtypes of breast cancer via intratumoural and peritumoural radiomic analysis with subregion identification based on the decomposition of contrast-enhanced magnetic resonance imaging (DCE-MRI).
研究成果
Radiomic analysis based on CAM decomposition of DCE-MRI signals effectively identifies intratumoural and peritumoural subregions, improving the prediction of breast cancer molecular subtypes. The fusion of predictive models from tumour and parenchymal subregions achieved the highest accuracy (AUC=0.897), suggesting clinical potential for noninvasive subtype assessment without biopsy. Future work should validate with additional datasets and imaging modalities.
研究不足
The image decomposition method requires high temporal resolution and is difficult with limited time-series data (less than three postcontrast series). The study did not include other imaging modalities like DW-MRI or T2-weighted imaging for validation. Tumour sizes were larger than 1 cm, and smaller tumours were not studied. The robustness of the model needs validation with different scanning protocols.
1：Experimental Design and Method Selection:
The study used a retrospective diagnostic methodology with a convex analysis of mixtures (CAM) method for unsupervised decomposition of DCE-MRI time-series signals to identify subregions in tumours and surrounding parenchyma. Radiomic features were extracted from these subregions, and a random forest model was employed for classification with leave-one-out cross-validation (LOOCV).
2：Sample Selection and Data Sources:
The dataset included 211 women with histopathologically confirmed invasive breast cancer, collected between January 2013 and July 2014. Patients with incomplete MR data, no immunohistochemical data, prior treatments, or no obvious tumour lesions were excluded.
3：Patients with incomplete MR data, no immunohistochemical data, prior treatments, or no obvious tumour lesions were excluded.
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: A 3.0-T MRI system (Siemens Medical Systems) with an eight-channel breast array coil was used. Gadobutrol contrast agent was injected intravenously. MATLAB software (2015a, MathWorks Inc.) was used for data analysis.
4：0-T MRI system (Siemens Medical Systems) with an eight-channel breast array coil was used. Gadobutrol contrast agent was injected intravenously. MATLAB software (2015a, MathWorks Inc.) was used for data analysis.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: DCE-MRI was performed with one precontrast and five postcontrast acquisitions. Image preprocessing included tumour annotation, segmentation using spatial fuzzy C-means algorithm, and registration. CAM decomposition was applied to identify subregions (plasma input, fast-flow kinetics, slow-flow kinetics). Radiomic features (statistical, texture, morphological) were extracted, and a random forest model was trained and tested with LOOCV for subtype prediction.
5：Data Analysis Methods:
Statistical analysis included χ2 tests, Fisher's exact test, ANOVA, Kruskal-Wallis test with Bonferroni correction. Random forest classification with nested LOOCV was used, and performance was evaluated using AUC from ROC curves.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
