(Sunday, 7/29/2018) 4:00 PM - 5:00 PM
Room: Exhibit Hall | Forum 1
Purpose: To investigate the impact of the size of the secondary ROI on statistical chi-square analysis of heterogeneity of liver tissue texture on ultrasound images.
Methods: This retrospective study was IRB approved. Ultrasound images of right lower lobe of thirty-one patients (22 M, 9 F; 17-81 y; F0-3: 16; F4-6: 15) who underwent non-targeted ultrasound guided biopsy were analyzed using an in-house Matlab program. A user manually selected a primary ROI in the range of 4-6 cm (avoiding non-hepatic parenchyma). Local variance, σ², of several texture features (image intensity, image gradient) were calculated in NxN pixels secondary ROIs in a roster scan fashion. The primary ROI contains a large number of secondary ROIs to maintain the precision of statistical analysis. The global variances, σ₀², of the texture features were estimated theoretically from local mean intensities and gradients. The χ²(=σ²/σ₀²) were calculated and plotted on a histogram. Several statistical descriptors were extracted from the histogram and entered in statistical analysis.This abstract investigated if the size of the secondary ROI affects the sensitivity and specificity of the χ² test. Three different size--5x5, 25x25, and 45x45--correspond to liver textures of 1mm, 5mm, and 9mm spatial scales.
Results: For image gradient, the performances of 25x25 and 45x45 ROIs in distinguishing advanced fibrosis (F4-6) from low grade fibrosis (F0-3) were comparable (25x25: p<0.001, AUC=0.985, sensitivity=93.7%, specificity=93.3%; 45x45: p<0.001, AUC=0.985, sensitivity=87.5%, specificity=100%). Both outperformed 5x5 pixels ROI (p=0.003, AUC=0.742, sensitivity=75.0%, specificity=73.0%). For image variance (noise), the performances of 25x25 and 45x45 ROIs were also comparable (25x25: p<0.001, AUC=0.813, sensitivity=62.5%, specificity=93.3%; 45x45: p=0.009, AUC=0.738, sensitivity=87.5%, specificity=53.3%). However 5x5 ROI outperformed them both (p<0.001, AUC=0.879, sensitivity=87.5%, specificity=80.0%).
Conclusion: The size of the secondary ROI substantially affects the sensitivity and specificity of THA. We recommend N≥25 and N=5 for analysis based on image gradient and image variance, respectively.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
