Hebei General Hospital researchers develop a radiomics-based machine learning pipeline to preoperatively predict spread through air spaces (STAS) in lung adenocarcinoma. They segment tumor regions on CT images, extract quantitative texture, shape, and intensity features, and apply LASSO and classifiers including a ResNet50 deep learning network. The model achieves AUCs up to 0.918, offering a non-invasive tool to guide surgical planning.
Key points
Extracted 3D CT radiomic features (texture, shape, intensity) screened via Mann–Whitney U, Spearman filtering, and LASSO reduction.
Combined clinical markers (CEA level, FEV1/FVC) with radiomics in a nomogram achieving AUC 0.878 for STAS prediction.
Employed ResNet50-based deep learning to derive 2D features, boosting classification AUC to 0.918 in machine learning models.
Why it matters:
This AI-driven radiomics approach enables non-invasive, accurate preoperative risk stratification for lung adenocarcinoma, improving surgical planning and patient outcomes.
Q&A
What is spread through air spaces (STAS)?
How does radiomics differ from traditional imaging?
Why use LASSO regression for feature selection?
What role does ResNet50 play in this study?
Read full article
Academy
Radiomics: Transforming Medical Images into Quantitative Data
Radiomics is an emerging field that converts standard medical images—such as CT, MRI, and PET scans—into high-dimensional, mineable data. Instead of visually interpreting image characteristics, radiomics uses computational algorithms to extract quantitative features describing tumor shape, intensity distribution, texture patterns, and spatial relationships. These features can reveal subtle differences in tissue heterogeneity, vascularity, and cellular organization that are not detectable by the human eye.
In practice, radiomics follows a pipeline:
- Image Acquisition: High-resolution images are obtained using standardized protocols to ensure consistency.
- Segmentation: Regions of interest (ROIs), such as tumors or organs, are outlined manually or via semi-automated tools.
- Feature Extraction: Hundreds to thousands of features, including first-order intensity statistics, shape descriptors, and texture matrices (e.g., GLCM, GLRLM, GLSZM), are computed.
- Feature Selection and Modeling: Statistical tests and machine learning methods (e.g., LASSO, random forests) identify the most predictive features and build models for diagnosis, prognosis, or treatment response.
Radiomics holds great promise for longevity research by enabling non-invasive biomarkers of age-related changes. For example, radiomic features of brain MRI can detect microstructural alterations in white matter that correlate with cognitive decline, while texture analysis of vascular imaging may uncover early signs of atherosclerosis contributing to cardiovascular aging.
Deep Learning for Biomedical Imaging in Longevity Research
Deep learning, particularly convolutional neural networks (CNNs), has revolutionized medical image analysis by automatically learning hierarchical representations from raw pixel data. Unlike handcrafted radiomic features, CNNs discover optimal filters during training, capturing complex patterns across multiple scales. A popular backbone, ResNet50, features 50 layers with residual connections that ease gradient flow and enable deeper architectures.
Applications in longevity science include:
- Brain Age Estimation: CNNs trained on large MRI datasets predict biological brain age, which correlates with cognitive performance and mortality risk.
- Cardiovascular Assessment: Deep models analyze echocardiograms and CT angiograms to quantify vessel stiffness, plaque burden, and early signs of arterial aging.
- Organ Biomarkers: Automated segmentation and feature learning in abdominal CT can assess liver fat accumulation and kidney morphology changes linked to metabolic syndrome.
By integrating radiomic and deep learning approaches, researchers can develop robust, non-invasive biomarkers for healthy aging, monitor intervention effects, and personalize longevity strategies. Combining quantitative imaging with AI accelerates discovery of early aging signatures and guides preventive measures to extend healthspan.
