Spatial process decomposition for quantitative imaging biomarkers using multiple images of varying shapes

Area of Science:

• Radiomics and Medical Imaging Analysis
• Statistical Modeling in Healthcare
• Biomarker Discovery

Background:

• Quantitative imaging biomarkers (QIB) are crucial for noninvasive disease detection, cancer monitoring, and precision medicine.
• Current QIB extraction methods are often ad hoc and lack reproducibility, hindering reliable analysis.
• There is a need for robust and generalizable statistical approaches in radiomics.

Purpose of the Study:

• To propose a general and flexible statistical framework for reproducible QIB extraction from 3D medical images.
• To develop a model-based spatial process decomposition for capturing relevant image features.
• To enable optimal prediction of underlying true images for accurate feature extraction.

Main Methods:

• A novel model-based spatial process decomposition is developed for 3D medical image analysis.
• Patient-specific random weights are utilized with common component functions for robust modeling.
• Maximum likelihood estimation is employed for model fitting and selection.
• Optimal prediction of the underlying true image is used for feature extraction.

Main Results:

• Simulation studies demonstrate the favorable properties of the proposed methodology.
• The approach effectively handles three-dimensional medical image data.
• Extracted QIBs show association with a clinical endpoint in a cancer image dataset.

Conclusions:

• The proposed statistical approach offers a reproducible and flexible method for QIB extraction in radiomics.
• This methodology can advance noninvasive disease detection, cancer monitoring, and precision medicine.
• The findings support the use of advanced statistical modeling for improved medical image analysis.