Purpose: Quantitative ultrasound imaging using texture analysis has advanced considerably in the past two decades. However, one of the main challenges for ultrasound texture analysis is the dependency of radiomics texture features on the gain-settings used during ultrasound scanning. The purpose of this study is to develop a mathematical model, specifically an analytical grayscale transformation (AGT) method, to systematically calibrate for the various gain-settings.
Methods: The proposed AGT approach utilizes a general-purpose phantom. The gray-level mean and distribution are quantified for the region of interest (ROI) for all scans. The cumulative probability distribution (CDF) of each ROI is fitted to a sigmoid function with R-squared and Root Mean Square Error (RMSE), within which the two parameters of the sigmoid function are monotonic to gain-setting that can be modeled using a Langevin equation. B-mode images of parotid glands (5 healthy volunteers and 5 head-and-neck cancer patients) were used to test the AGT method. To evaluate the performance of our gain-setting calibration method in radiomics study, radiomics texture features were calculated using matrices including the gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level size zone matrix (GLSZM) and neighborhood gray-tone difference matrix (NGTDM). The relative standard deviation was calculated for images before and post calibration.
Results: In this study, CDF of each ROI was fitted to a sigmoid function with R-squired >0.98 and RMSE <0.005. With the AGT method, the relative standard deviation of gain-setting induced variation in major texture features was reduced to less than 5%.
Conclusion: We have developed an AGT calibration method that can significantly reduce gain-setting induced variation for radiomics feature analysis.