Purpose: Evaluation of regional hepatic perfusion by contrast-enhanced ultrasound (CEUS) is helpful to the differential diagnosis of focal liver lesions (FLLs). Because most patients cannot hold their breath for the duration of the entire hepatic perfusion scan, ultrasonographists tend to employ free-breathing acquisition method. A new strategy using a combination of template matching and frame selection is proposed to correct the respiratory motion and improve the accuracy of the quantification evaluation of the perfusion.
Methods: Considering that most commercial ultrasound machines can provide a dual display mode for simultaneously visualizing contrast and tissue images, the registration of the contrast images was achieved by the registration of the corresponding tissue images. After the template was located, the rough search space was estimated using a priori knowledge of the tumor location in the free-breathing image sequence. Then, a simple double-selection method was proposed to select the similar images from a large number of successive matched images via global and local threshold settings. In this method, alpha and m were the offset of the global threshold and the time interval for setting local sampling range, respectively. These two parameters were also investigated. The strategy was tested on ten liver CEUS acquisitions with a handle probe by using sum of absolute differences (SAD) metric. The visual evaluation for 2D image sequences and the extracted time-intensity curves from the regions of interest were performed. Simpler curve descriptors of the motion-uncorrected and motion-corrected image sequences were calculated on a pixel-by-pixel basis and evaluated as parametric perfusion maps. The quality of these parametric images was compared, in terms of both the accuracy and spatial resolution. For the corrected and uncorrected sequences, their mean deviation values (mDVs) and mean quality-of-fits (mQOFs) were measured.
Results: When alpha and m were both set to 0.5, 9.20 +/- 3.22% of the total number of frames were selected. After the motion correction, the mDVs of all the image sequences decreased from 21.69 +/- 2.80 to 13.78 +/- 2.68. The mQOFs of all the corrected sequences increased by an average of 15.32 +/- 5.13%. The quality of curve fitting and the corresponding parametric imaging computed on motion-corrected sequences were improved. On the average, the motion correction of a sequence containing about 100 frames was performed in approximately 3 min using MATLAB, whereas a completely manual approach requires approximately 10 min.
Conclusions: The image-based strategy independent of the tumor size can quickly correct respiratory motion in CEUS image sequences. Simple manual operation is only needed, such as the selection of the template image and search space. It is user-friendly and suitable for most clinical cases affected by adverse factors of sampling. Due to the merit of the saving time, this strategy can be widely applied to clinical practice, and the diagnostic efficiency of FLLs will be improved. Moreover, the correction strategy is a key preprocessing step toward local quantification of hepatic perfusion studies.