Related Experiment Video
Updated: Aug 6, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
Physics-Informed Multiscale Decoding of Tissue Microstructure: The Gray Level Affinity Metrics (GLAM) Framework
Ahmad Pour Rashidi1, Laetitia Perronne1, Chase Krumpelman1
1Department of Radiology, Northwestern University Feinberg School of Medicine, Chicago, IL, USA.
Journal of Imaging Informatics in Medicine
|July 20, 2026
Summary
Physics-inspired Gray Level Affinity Metrics (GLAM) offer advanced texture analysis for high-grade glioma. GLAM captures multiscale tumor architecture, improving prognostic accuracy and risk stratification in diverse molecular subtypes.
Area of Science:
- Radiomics and Medical Imaging
- Statistical Physics in Biology
- Computational Pathology
Background:
- Conventional radiomics uses localized statistics, potentially missing broader structural information in medical images.
- Statistical mechanics offers a framework to link microscopic interactions to macroscopic properties, applicable to image texture analysis.
Purpose of the Study:
- Introduce Gray Level Affinity Metrics (GLAM), a novel physics-inspired texture descriptor family.
- Evaluate GLAM's ability to capture multiscale tumor architecture and improve prognostic performance in high-grade gliomas.
- Compare GLAM with conventional radiomics for precision prognostication across different molecular subtypes.
Main Methods:
- Developed GLAM using Radial Distribution Functions, treating image voxels as interacting particles.
- Quantitatively assessed GLAM's intrinsic dimensionality and information content compared to standard texture metrics.
- Validated prognostic models using Leave-One-Center-Out cross-validation and bootstrap iterations on a multi-center high-grade glioma cohort.
Main Results:
- GLAM demonstrated significantly higher intrinsic dimensionality than standard texture metrics, capturing substantial non-redundant information.
- Standalone GLAM achieved optimal prognostic performance in the MGMT promoter-methylated subgroup (Mean Test C-Index 0.646).
- A combined GLAM and radiomics model showed strong performance in the MGMT promoter-unmethylated subgroup (Mean Test C-Index 0.643) with high cross-institutional stability.
Conclusions:
- GLAM effectively characterizes overarching multiscale tumor architecture using statistical physics principles.
- The synergy between GLAM and conventional radiomics enhances precision prognostication across diverse molecular glioma subtypes.
- GLAM represents a significant advancement in texture analysis for medical imaging, offering improved domain resistance and prognostic capabilities.
