The Markusovszky University Teaching Hospital team employs machine learning to analyze pre-treatment CT-derived Hounsfield unit statistics and lung volume data, training decision trees, kernel-based classifiers, and k-nearest neighbors to predict patients at risk of radiation-induced lung fibrosis following breast radiotherapy, supporting personalized treatment planning.
Key points
Extracted CT lung density metrics (HU mean, SD, min, max) and lung volume from planning scans.
Trained Fine Tree, optimizable kernel, and kNN models with five-fold cross-validation on 242 breast radiotherapy cases.
Developed a simple HPF score combining HU metrics and lung volume achieving 62.8% validation accuracy for RILI risk.
Why it matters:
This approach enables proactive identification of patients at high risk for radiation-induced lung fibrosis, improving treatment personalization and reducing pulmonary toxicity.
Q&A
What are Hounsfield units?
How does the Human Predictive Factor (HPF) work?
Why use multiple ML models instead of one?
What are the main limitations of this study?
Read full article
Academy
Pulmonary Fibrosis and Healthy Lung Aging
Pulmonary fibrosis is a condition where lung tissue becomes damaged and scarred, making it harder to breathe and deliver oxygen to the bloodstream. Over time, scar tissue replaces healthy lung cells, reducing lung elasticity and gas exchange capacity. In the context of longevity and healthy aging, preserving lung function is crucial because respiratory health impacts overall energy levels, physical activity, and quality of life.
Key factors in pulmonary fibrosis include:
- Inflammation: Persistent lung inflammation can trigger fibroblast activation and collagen deposition.
- Environmental Exposures: Pollutants, radiation, and certain drugs can harm lung tissue.
- Genetic Susceptibility: Some individuals inherit a predisposition to abnormal wound healing.
Preventing or slowing fibrosis involves anti-inflammatory strategies, antioxidant support, and therapies that target fibrotic pathways such as TGF-β signaling. Maintaining lung health through regular exercise, avoiding pollutants, and monitoring occupational exposures supports healthy aging.
Radiomics in Medical Imaging
Radiomics is an emerging field that extracts quantitative features from medical images, such as CT or MRI scans, to reveal patterns beyond human perception. By converting imaging data into high-dimensional mineable information, radiomics enables clinicians and researchers to uncover biomarkers for diagnosis, prognosis, and treatment response.
Components of radiomics:
- Image Acquisition: High-resolution scans with standardized protocols ensure reproducible data.
- Segmentation: Defining regions of interest—such as healthy lung tissue versus fibrotic areas—using manual or automated methods.
- Feature Extraction: Computing metrics like texture, intensity histograms, shape descriptors, and wavelet transforms to characterize tissue heterogeneity.
- Data Analysis: Applying statistical models and machine learning algorithms to correlate radiomic features with clinical outcomes.
In longevity science, radiomics can identify early signs of organ deterioration, monitor disease progression, and guide interventions to preserve function. For lung health, radiomics features like Hounsfield unit variations and texture complexity can serve as noninvasive biomarkers for fibrosis risk, enabling earlier therapeutic strategies that support healthy aging.
