Precision Radiology: Predicting longevity using feature engineering and deep learning methods in a radiomics...

Area of Science:

• Radiomics and Medical Imaging Analysis
• Precision Medicine
• Biomarker Discovery

Background:

• Precision medicine requires non-invasive methods to assess individual health status, combining genetic and environmental factors.
• Current limitations in phenotypic variation assessment hinder early intervention and chronic disease management.
• Accurate health status knowledge is crucial for personalized treatment decisions.

Purpose of the Study:

• To demonstrate the use of routine CT imaging for predicting patient longevity using computer image analysis.
• To explore radiomics techniques for extracting health-related biomarkers from medical images.
• To assess the potential of deep learning in radiomics research for precision medicine.

Main Methods:

• Utilized computer image analysis techniques on routinely acquired cross-sectional CT scans.
• Applied machine learning methods, including convolutional neural networks, to radiomics data.
• Validated longevity predictions against established clinical methods.

Main Results:

• Proof-of-concept experiments demonstrated the feasibility of predicting patient longevity from CT images.
• Radiomics techniques successfully extracted biomarkers relevant to mortality prediction.
• The developed computer image analysis methods achieved results comparable to manual clinical methods.

Conclusions:

• Routine CT imaging, analyzed with radiomics and deep learning, can serve as a non-invasive tool for health assessment.
• This approach offers substantial potential for advancing precision medicine initiatives.
• Computer image analysis of medical images can significantly enhance biomarker discovery and patient outcome prediction.