Related Experiment Video
Updated: Aug 10, 2025
05:45
Quantitative Immunohistochemistry of the Cellular Microenvironment in Patient Glioblastoma Resections
Published on: July 31, 2017
9.7K
MR Intensity Normalization Methods Impact Sequence Specific Radiomics Prognostic Model Performance in Primary and
Patrick Salome1,2,3,4, Francesco Sforazzini1,2,3, Gianluca Grugnara5
1Clinical Cooperation Unit (CCU) Radiation Oncology, German Cancer Research Centre, INF 280, 69120 Heidelberg, Germany.
Cancers
|February 11, 2023
Summary
Intensity normalization methods significantly impact radiomics models for predicting high-grade glioma survival. Performance varies by MRI sequence, highlighting the need for sequence-specific approaches in survival analysis.
Area of Science:
- Radiomics
- Medical Imaging Analysis
- Oncology
Background:
- Radiomics models extract quantitative features from medical images to predict clinical outcomes.
- Intensity normalization (IN) is a crucial preprocessing step in radiomics, aiming to reduce image variability.
- The impact of different IN methods on survival prediction models for gliomas remains incompletely understood.
Purpose of the Study:
- To investigate how various intensity normalization (IN) methods affect the performance of overall survival (OS) radiomics models.
- To evaluate these impacts across different magnetic resonance (MR) imaging sequences for primary (pHGG) and recurrent high-grade glioma (rHGG).
Main Methods:
- Retrieved MR scans from two independent cohorts (rHGG, pHGG) acquired before radiotherapy.
- Extracted sequence-specific significant features (SF) associated with OS from tumor volumes after applying 15 IN methods.
- Utilized Cox proportional hazard and Poisson regression models for survival analysis, ranking methods by 10-fold cross-validated concordance index (C-I), mean square error (MSE), and Akaike information criterion (AIC).
Main Results:
- Intensity normalization methods demonstrated a significant impact on survival prediction performance, with varying effects across MR sequences (C1/C2 C-I range: 0.62-0.71/0.61-0.72).
- White stripe normalization showed stable results for T1wce, while Combat and histogram matching (HM) performed consistently for T2w and T1w sequences in one cohort.
- Eliminating IN-impacted features led to a mean decrease of 0.05 in C-I and 0.03 in MSE across all sequences.
Conclusions:
- The choice of intensity normalization method critically influences the predictive power of survival models in glioma.
- Radiomics model performance is dependent on the specific MR imaging sequence used.
- Sequence-specific optimization of IN methods is recommended for robust survival prediction in high-grade gliomas.
