Premium
SU‐F‐R‐15: Establishing Relevant ADC‐Based Texture Analysis Metrics for Quantifying Early Treatment‐Induced Changes in Head and Neck Squamous Cell Carcinoma
Author(s) -
Loman K,
Nawrocki J,
Hoang J,
Yoo D,
Chang Z,
Mowery Y,
LI X,
Peterson B,
Brizel D,
Craciunescu O
Publication year - 2016
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.4955787
Subject(s) - wilcoxon signed rank test , effective diffusion coefficient , head and neck squamous cell carcinoma , medicine , nuclear medicine , histogram , voxel , head and neck cancer , diffusion mri , head and neck , mathematics , radiology , artificial intelligence , magnetic resonance imaging , mann–whitney u test , computer science , radiation therapy , surgery , image (mathematics)
Purpose: The purpose of this study is to identify texture analysis metrics from apparent diffusion coefficient (ADC) maps that would provide quantifiable changes with treatment in patients with head and neck squamous cell carcinoma (HNSCC). We discerned which imaging metrics were relevant using baseline agreement and variations during early treatment. Methods: We retrospectively analyzed diffusion‐weighted MRI scans in 9 patients with stages II‐IV HNSCC. ADC maps were generated from two baseline scans, performed 1 week apart, and one early treatment scan, obtained during the 2nd week of chemoradiation. Regions of interest (ROI) consisting of primary and nodal disease were drawn on the resampled ADC maps. Four 3D texture matrices describing local and regional relationships between voxel intensities in the ROIs were generated. From these, 38 texture metrics and 7 histogram features were calculated for each patient, including the mean and median ADC. To identify metrics with good agreement between baseline studies, we compared all metrics using the intra‐class correlation coefficient (ICC). For metrics with ICC ≥ 0.80, the Wilcoxon signed‐rank test was used to test if the difference between the mean of the baselines and the early treatment was non‐zero. Results: Nine of the 45 metrics had an ICC ≥ 0.80. Six of these 9 metrics had a p‐value < 0.05: run length non‐uniformity, ADC median, texture strength, ADC mean, zone percentage, and variance. Only 1 of the 9 metrics remained of interest, after applying the Holm correction to the alpha levels: run length non‐uniformity (p = 0.004) in the Gray Level Run Length Matrix. Conclusion: The feasibility of texture analysis is dependent on the baseline agreement of each metric, which disqualifies many texture characteristics. Consequently, only a few metrics are reproducible and qualify for future studies that provide quantitative assessment of early treatment changes for HNSCC.