Premium
SU‐F‐I‐45: An Automated Technique to Measure Image Contrast in Clinical CT Images
Author(s) -
Sanders J,
Abadi E,
Meng B,
Samei E
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 - 10.1118/1.4955873
Subject(s) - hounsfield scale , histogram , percentile , standard deviation , segmentation , contrast (vision) , artificial intelligence , nuclear medicine , image processing , computer science , pattern recognition (psychology) , medicine , computed tomography , mathematics , radiology , image (mathematics) , statistics
Purpose: To develop and validate an automated technique for measuring image contrast in chest computed tomography (CT) exams. Methods: An automated computer algorithm was developed to measure the distribution of Hounsfield units (HUs) inside four major organs: the lungs, liver, aorta, and bones. These organs were first segmented or identified using computer vision and image processing techniques. Regions of interest (ROIs) were automatically placed inside the lungs, liver, and aorta and histograms of the HUs inside the ROIs were constructed. The mean and standard deviation of each histogram were computed for each CT dataset. Comparison of the mean and standard deviation of the HUs in the different organs provides different contrast values. The ROI for the bones is simply the segmentation mask of the bones. Since the histogram for bones does not follow a Gaussian distribution, the 25th and 75th percentile were computed instead of the mean. The sensitivity and accuracy of the algorithm was investigated by comparing the automated measurements with manual measurements. Fifteen contrast enhanced and fifteen non‐contrast enhanced chest CT clinical datasets were examined in the validation procedure. Results: The algorithm successfully measured the histograms of the four organs in both contrast and non‐contrast enhanced chest CT exams. The automated measurements were in agreement with manual measurements. The algorithm has sufficient sensitivity as indicated by the near unity slope of the automated versus manual measurement plots. Furthermore, the algorithm has sufficient accuracy as indicated by the high coefficient of determination, R2, values ranging from 0.879 to 0.998. Conclusion: Patient‐specific image contrast can be measured from clinical datasets. The algorithm can be run on both contrast enhanced and non‐enhanced clinical datasets. The method can be applied to automatically assess the contrast characteristics of clinical chest CT images and quantify dependencies that may not be captured in phantom data.