A Visually Interpretable, Dictionary-Based Approach to Imaging-Genomic Modeling, With Low-Grade Glioma as a Case...

Area of Science:

• Medical imaging
• Radiomics
• Computational pathology

Background:

• Radiomics extracts imaging features for clinical outcome prediction.
• Current radiomic models lack visual interpretability, hindering clinical adoption.
• Intratumor heterogeneity is challenging to capture with traditional biopsies.

Purpose of the Study:

• To develop visually interpretable radiomic features using dictionary learning.
• To predict genetic alterations in low-grade gliomas.
• To associate interpretable features with molecular pathways in gliomagenesis.

Main Methods:

• Applied dictionary learning to extract features from MRI scans of low-grade gliomas.
• Utilized a publicly available dataset for model training and validation.
• Correlated derived visual features (atoms) with known biomarkers (1p/19q codeletion, IDH1 mutation).

Main Results:

• The dictionary learning model accurately predicted 1p/19q codeletion and IDH1 mutation status.
• Identified specific image regions (atoms) associated with these genetic biomarkers.
• Demonstrated that these regions correlate with key gliomagenesis molecular pathways.

Conclusions:

• Dictionary learning offers a promising approach for enhancing radiomic model interpretability.
• This method can provide insights into the diagnostic process for low-grade gliomas.
• Potential to assist radiologists in selecting optimal biopsy locations for molecular analysis.