Graphical flow chart illustrating the different stages of FFMCT. The main concepts include using a patient based model to define an image quality plan, then seeking modulated fluence fields that aim to achieve the prescribed plan, and finally acquiring the modulated set of projections and reconstructing the FFMCT image. The dashed ellipses identify hypothetical regions of prescribed high image quality (low noise in this case) data. However, the image quality plan could define one or multiple irregularly shaped regions of interest with prescribed image quality that varies regionally, depending on the task; the plan could also include one or more metrics (e.g., contrast-to-noise ratio, dose, etc.).
(a) Illustration of the simulated anthropomorphic chest phantom used in this study. Prescribed image quality distributions, where white is equivalent to high image quality, are shown for the cases where the scanning priority is (b) the entire patient cross-section, (c) the lung, and (d) the heart. The associated clinically relevant region of interest delineated on image (a) is shown in the upper left hand corner of images (b)–(d).
Modulation profiles (upper rows) as functions of detector position (ξ) and projection angle (θ) under different modulation constraints with each resulting IQMsd (lower rows) shown in Hounsfield units for the case of (a) diagnostic, (b) lung-screening, and (c) cardiac CT scans. The rightmost column indicates the prescribed standard deviation for the corresponding imaging application and the IQMsd for a reference uniform fluence field. Upper left hand corner of each prescribed IQMsd shows the clinically relevant region of interest for each scenario overlain on the illustration of the anthropomorphic chest slice.
Illustration of the overlap between the collimated exposure and the prioritized ROI for the lung optimized, two exposure case. The gray region depicts where the lungs are projected onto the detector, while the white region indicates overlap of the lungs with the collimated exposure. In many instances, the collimated field of view is substantially less than the ROI.
Mean image quality within the prioritized regions of interest in terms of the average voxel standard deviation for different constraints in FFMCT and for different imaging tasks (i.e., diagnostic, lung, and heart scans) and a uniform image quality weighting. The target standard deviation value was 10 HU. Error bars represent two standard deviations and provide a measure of the uniformity of the image quality (i.e., image noise), where larger error bars indicate less uniformity. The case of the uniform fluence field is included for reference, showing the least uniformity in image quality for all regions of interest.
Standard deviation-volume histograms (SDVHs) for the (a) routine diagnostic CT, (b) lung-screening, and (c) cardiac CT cases over the corresponding regions of interest and for a uniform image quality weighting. The vertical axis represents the percent of volume that has a standard deviation equal to or greater than a given standard deviation in HU along the horizontal axis. Broader curves represent less uniformity in image quality (i.e., noise). The ideal curve would be a step function at the target of 10 HU, indicating that 100% of voxels in the volume have a standard deviation of 10 HU.
Average root mean square error between the predicted outcomes and the prescribed values across thoracic imaging applications for different constraints in FFMCT and for the uniform image quality weighting priority. The break in the vertical axis marks a change in value and scale. The errors corresponding to the prioritized regions of interest are approximately an order of magnitude less than the total error.
(a) High resolution reconstructions produced using projection data with added simulated Poisson noise for different modulation profiles (see corresponding modulation profiles in Fig. 3 ). Top left corner of each image shows the region of interest of clinical relevance. A simulated lesion with 3% signal deviation from the mean soft tissue value is added to the lung scenario (see boxed region). The display scale was narrowed to highlight the different noise manifestations between image reconstructions. (b) Difference images between the reconstructions in (a) above and a reference noiseless reconstruction image. Top left corner of each image shows a representation of the predicted IQMsd. The noise distributions of the images in (b) can be seen to follow the same trend as the distribution within the predicted IQMsd. A comparison between simulated results and the predicted values was performed on the region contained within the dashed ellipse for validation of the noise prediction model.
Enlarged image of boxed region in Fig. 8 . The arrow points to a simulated lesion with a 3% signal deviation from the mean soft tissue value of the anthropomorphic phantom. Streak artifacts to the right of the image correspond to the defined lower image quality region. Streaks are a result of spatial nonuniformity in the image noise (i.e., directional noise correlations). The lesion is visible within the high image quality region but would not be visible within regions of the low image quality region of interest due to the increased noise content.
Relative integral dose normalized to the case of a uniform fluence field that would provide the same average image quality as the target prescribed value. For additional reference, the dashed line shows the integral dose of a bowtie filter with uniform tube current that would meet the same minimum image quality as the diagnostic, bowtie constrained case.
SDVH curves for the routine diagnostic scan when the cost function is weighted to prioritize achieving a minimum prescribed image quality rather than image quality uniformity. Relative to Fig. 6 , the static filter cases (i.e., the uniform, bowtie, and custom filter cases) present SDVH curves that are similar except for a significant horizontal shift to accommodate the requirement of not exceeding the maximum standard deviation. The shapes of the remaining curves become increasingly more asymptotic near the maximum prescription point.
Relative integral doses for different constrained modulation cases and imaging tasks when the cost function is weighted to prioritize achieving a minimum image quality prescription. The greater variation in relative dose between fixed filter and dynamic modulation cases reflects the larger difference between relative average image qualities.
Article metrics loading...
Full text loading...