A novel saline infusion sonohysterography-based strain imaging approach for evaluation of uterine abnormalities in vivo: preliminary results.

Figure 2 presents a B-mode image ( Figure 2A ) and the corresponding strain image ( Figure 2B ) estimated during saline infusion sonohysterography in a 50-year-old patient who later had a diagnosis of a fibroid based on the clinical sonographic examination. The red arrows on both the B-mode and strain images indicate the location of the fibroid within the uterine cavity. The yellow contours indicate the outer uterine wall. Clinical diagnosis is based on the radiologist’s reading of the B-mode image loop obtained during the diagnostic saline infusion sonohysterographic procedure. The strain image provides a better definition of the fibroid, with a large strain boundary surrounding the fibroid, referred to as the halo, and a low strain region (darker) within the outline of the fibroid. The fibroid is outlined with the characteristic decorrelation halo that was also observed as a slipping artifact during the ex vivo imaging of fibroids previously described by Hobson et al. 12 Both the location and area of the fibroid highlighted along the edges are clearly visualized on the strain image in Figure 2B . Figure 3A shows a B-mode image from a 46-year-old patient who had a diagnosis of a polyp based on diagnostic saline infusion sonohysterography, and the corresponding strain image is shown in Figure 3B . In a similar manner as described for Figure 2 , the multilevel hybrid technique is used to estimate the local displacements and to generate the strain image ( Figure 3B ). The red arrows on the B-mode and strain images indicate the location of the uterine polyp. The strain image clearly shows the presence and location of the uterine polyp when compared to the B-mode image. This patient underwent a hysterectomy based on the clinical sonographic findings obtained during the saline infusion sonohysterographic procedure, and the pathology report was based on the excised uterus. The polyp appears to be brighter or as a region with higher strain when compared to the background myometrium, indicating that the polyp is softer than the surrounding tissue. Figure 4A shows a B-mode image from a 61-year-old patient whose pathology report indicated the presence of endometrial cancer along with a fibroid. This patient also underwent a hysterectomy based on the clinical sonographic findings obtained during saline infusion sonohysterography, and the pathology report was based on the excised uterus. Observe that it is very difficult to visualize the cancer and fibroid on the B-mode image in Figure 4A . The uterine wall is outlined by the yellow contour on the B-mode image, and the red arrow indicates the region of interest. Note that on the strain image in Figure 4B , we see a highlighted region enclosing a stiffer region, indicating the location of the fibroid with the characteristic decorrelation halo. The region to the left of the fibroid is less uniform, which may be due to the presence of the diffuse endometrial cancer. Finally, Figure 5A shows a B-mode image from a 44-year-old patient that does not indicate the presence of masses; however, after processing the radiofrequency data, the strain image in Figure 5B shows the possibility of the presence of a mass within the uterine cavity.

The quality of the strain image obtained is dependent on the amount of deformation induced by the saline injection into the uterine cavity. If the amount of saline injected is too low, the deformation induced is too small to provide strain images with a reasonable signal-to-noise ratio. This situation has led to generation of suboptimal strain images in some instances because of the minimal deformation of the tissue. The amount of deformation applied via saline injection is dependent on the ability of the patient to tolerate the discomfort induced due to saline injection into the uterus; in other cases, insufficient deformation is due to the presence of saline from a previous infusion and the inability to withdraw sufficient fluid before starting a new infusion. Patient comfort is of paramount importance during the data acquisition because saline injection can be quite painful and distressing to the patient. Care should be taken to induce a sufficient deformation of the uterine wall that is tolerated by the patient to obtain high-quality strain images as well as having no saline or a minimum amount of it before starting the research data collection process. In this article, we illustrate the application of saline infusion sonohysterography-based strain imaging to detect and image the variations in the stiffness of uterine masses in vivo. The results show a clear indication of the presence of masses in the uterine cavity when compared to the corresponding B-mode images. The fibroids appear darker on the strain images, whereas the polyps appear brighter because of the difference in stiffness between the mass and surrounding background myometrial tissue. We also show a case in which the B-mode image does not appear to differentiate any masses, but the strain image indicates the possible presence of a uterine mass. However, on the basis of the limited data presented in this article, saline infusion sonohysterography-based strain imaging is expected to function only as a supplementary or adjunct imaging modality for characterization of focal uterine masses diagnosed on B-mode images. The use of saline infusion sonohysterography-based strain imaging as a stand-alone imaging technique for the detection of uterine masses requires additional validation on a larger number of patients and at multiple clinical sites. Further investigation is also required to determine the in vivo contrast for different pathologic conditions. Color and spectral Doppler correlation in lesions of interest will be investigated in the future and might add value to the analysis. In this study, only a qualitative comparison of the underlying mass with the surrounding myometrial tissue was performed. Quantitative comparisons by solving the inverse problem or shear wave imaging approaches could be used in the future. 22 , 23

Data acquisition for saline infusion sonohysterography-based strain imaging is performed on adult female patients in the radiology sonography suites of the University of Wisconsin–Madison Hospital and Clinics. The patients recruited for the study had already been scheduled for clinical saline infusion sonohysterography. Patient consent for the research study was obtained before the procedure. The data acquisition protocol and research study were approved by the University of Wisconsin-Madison Institutional Review Board. Clinical saline infusion sonohysterography is first performed on the patient following the appropriate clinical standard of care. The procedure is generally performed early in the menstrual cycle when the menstrual period has stopped or almost stopped but before ovulation. The patient is placed in the traditional lithotomy position, and a baseline transvaginal sonographic examination is performed. Subsequently, the transducer is removed; a sterile speculum is inserted into the vagina; and the external os of the cervix is brought into view. The cervix is cleansed with a povidone-iodine solution and then cannulated with a 28-cm 5F balloon catheter. Once inside the uterine cavity, the balloon is inflated with 2 to 3 mL of saline to fix the catheter in place. The speculum is then removed, and the endovaginal probe is reinserted beside the catheter. Under sonographic observation, the balloon is retracted so it does not occlude the internal cervical os. The balloon is positioned either within the endocervical canal or in the lower uterine segment. Then, 10 to 30 mL of warmed saline is hand injected through the catheter at a constant slow rate. Although not objectively modulated, this rate is approximately 0.5 to 0.75 mL/s. The distension of the uterine cavity is monitored to ensure that it is gradually increasing, despite concomitant loss of fluid through the fallopian tubes. Such a slow rate tends not to induce pain, and indeed, pain was not a limiting factor in any of the cases. The rate of injection is not varied systematically. Sagittal and coronal images of the distended endometrial cavity are then obtained. Saline is then withdrawn from the uterine cavity through the catheter, and the ultrasound system is configured (placed in the research mode) to capture radiofrequency data loops. Saline is then reinfused, and ultrasonic radiofrequency data frames are recorded during the infusion procedure. Figure 1 presents a flowchart for the data acquisition procedure on the patient. At the end of the procedure, the catheter balloon is deflated and removed. The entire procedure takes about 30 to 60 minutes depending on insertion and securing of the catheter with the balloon inflated to avoid slipping or catheter movement. Radiofrequency data acquisition for research, however, takes only 5 to 10 minutes. Saline has to be withdrawn out of the uterine cavity, followed by a subsequent reinjection, to obtain sufficient deformation of the uterine cavity for strain imaging. B-mode sonograms and radiofrequency data are acquired with a Sonoline Antares system (Siemens Medical Solutions, Inc, Mountain View, CA) using an EC9-4 transvaginal array transducer. The transducer is excited at a center frequency of 8 MHz, and the radiofrequency data are acquired at a sampling frequency of 40 MHz. A single transmit focus is used during the imaging process with dynamic focusing during receive. The maximum number of frames that the system can acquire during injection of the saline is stored (typically 150 frames) for offline processing. Fourteen patients provided informed consent and participated in the research study. The mean age of the patients scanned was 44 years. Among the 14 patients, no masses were seen in 7; however, 3 had a diagnosis of fibroids, and 2 each had a diagnosis of polyps and endometrial cancer, respectively, based on pathologic results. The B-mode images obtained during saline infusion sonohysterography were read and evaluated by a radiologist and taken into consideration when evaluating the saline infusion sonohysterography-based strain images. The next section describes the data processing performed to obtain the strain images. We use a new hybrid 2D cross-correlation algorithm, 21 which uses multiple 1-dimensional cross-correlation processing steps to estimate local displacements, along with 2D surface fitting on a sector grid to obtain subsample displacement estimates. The predeformation and postdeformation radiofrequency data are determined from the radiofrequency data loop acquired during saline infusion sonohysterography. The first frame in the data loop is initially selected as the predeformation frame, with the postdeformation frame separated by a specified frame interval. A frame interval of 5 is used for the images shown in this article. However, the frame interval can be increased if the deformation between the predeformation and postdeformation frames is too small to estimate local displacements and to compute the strain distribution. We use the deformation of the uterine wall induced by the saline injection during saline infusion sonohysterography as the mechanical stimulus for the strain imaging. The strain estimation algorithm first estimates coarse displacements using the 1-dimensional normalized cross-correlation function on the predeformation and postdeformation envelope data frames. If the normalized correlation coefficient between the specified gated window and the corresponding neighboring gated A-line segment is lower than a threshold value, an interpolation step between the two neighboring radiofrequency segments of the postcompression data is performed, and the normalized cross-correlation function is recalculated as described by Chen and Varghese. 21 The peak of the cross-correlation function over several normalized crosscorrelation functions in the 2D matrix is located, and a 2D surface fit with sector geometry is used to obtain subsample displacement estimates along the beam direction. The slope of the local displacements estimated in this manner is computed to obtain the strain distribution. For all of the data processing described in this article, an overlap of 67% between rectangular gated windows is used with an 18-wavelength window length along the beam direction and 11 A-lines along the azimuthal direction; this procedure is done during the first processing step that uses envelope signals. Envelope signals are generated from the radiofrequency data using a Hilbert transform. The remainder of the processing is done with the radiofrequency echo signals to calculate accurate and precise displacements while progressively reducing the dimensions of the 2D processing window length along the beam direction and number of A-lines in the azimuthal direction. For the second processing step, a 75% overlap is used with a 12-wavelength window length along the beam direction and 11 A-lines along the azimuthal direction. The third and final processing step uses a 67% overlap with a 6-wavelength window length along the beam direction and 7 A-lines along the azimuthal direction.