Abstracts for the 2022 Symposium on Ultrasonic Imaging and Tissue Characterization

Abstract

The Field II ultrasound simulation developed by recently reached its 25-yearanniversary. In that time, its impact on the development of novel methods and systems for medical imaging is hard to overstate. This software has been made freely available to the ultrasound community as citation ware( > 2700 as of 2022) and is frequently updated to support modern versions of Matlab. I will provide a brief retrospec-tive on Field II including describing its simulation methods, capabilities, and limitations to put its use into context among a growing number of other simulation approaches for modern ultrasound research. This talk will highlight our group’s use of Field II in several areas of research to demonstrate how we leverage its linear simulation approach for fundamental acoustic studies. I will discuss best practices for simulation including generation of additive noise. I will demonstrate the combination of pre-computed targets for use in training machine learning applications. I will explore the use of the multistatic data set in the efficient creation and evaluation of various imaging sequences, especially for synthetic aperture imaging. Work from others that has been used to complement the capabilities of Field II will also be briefly introduced(e.g. introducing additive acoustic clutter models, generating imaging targets from natural images for machine learning, the use of simulated acoustic fields as input for mechanical simulations). sound speed Ultrasonic backscatter is associated to cardiac collagen deposition, while anisotropy in ultrasonic backscatter is associated with myo fiber alignment. Preliminary data from our lab suggested anisotropy in backscatter may be primarily associated with collagen that aligns parallel to myofibers, not the myofibers themselves. The purpose of the present study was to determine a relationship between myocardial collagen and anisotropy of ultrasonic backscatter in left ventricular short axis images. Hearts were excised from Sprague Dawley rats, aligned in the short axis with the anterior wall closest to the transducer, and perfused with a colla-genase-containing solution for either 10 (n=7) or 30 minutes (n=7)or control solution for 30 minutes(control n=8). Serial ultrasound images were acquired throughout collagenase digestion and ultrasonic backscatter was assessed where the collagen is primarily aligned perpendicular to the angle of insonification(anterior and posterior walls), and where collagen is primarily aligned parallel to the angle of insonification (lateral and septal walls). Our data suggested that collagenase digestion reduced backscatter anisotropy within the myocardium (p < 0.001)with the lateral and septal walls (collagen parallel to ultrasound) showing the greatest change in backscatter intensity. Histology (Trichrome staining) and biochemistry (hydroxyproline assay) suggests that collagen remains present but is crosslinking is altered within 10 minutes(p < 0.047). These data suggest the anisotropy of ultrasonic backscatter is largely influenced by myocardial collagen crosslinking. Multiple angles of insonification may provide quantifiable indexes for both alignment (fiber orien-tation)and of the status of collagen-dominated × 10 -4 and 7.48 × 10 -5 – 2.20 × 10 -4 , respectively, depending on the specimen and transducer frequency. Measured values for the spatial mean and standard deviation of the acoustic impedance ranged between 1.49 – 1.51 MRayls and 0.011 – 0.024 MRayls, respectively. Detailed, two-dimensional maps of acoustic impedance and reflection coefficient were produced, providing a clear visualization of the spatial variation of these ultrasonic properties of normal mammalian brain. Elastography is a quantitative ultrasound (US) technique used to obtain soft tissue stiffness maps. This technique is routinely used in clinic by using the shear wave elastography (SWE) method where shear waves (SW) are commonly generated by acoustic radiation force at low-medium US frequency (between 3 and 15 MHz). In this work, a mechanical vibrator, coupled to an ultrafast and US high frequency device (Vevo F2, Visualsonics), is used to generate shear wave. It shows the capability of this device to catch propagation of SW, leading to strong improvement in the spatial resolution of stiffness maps at US frequencies higher than 15 MHz. Experiments were performed on a calibrated tissue-mim-icking phantom, in which transient SW were generated by an external mechanical vibrator, using transient sinusoidal from 200 to 600 Hz. The vibrator is triggered by the Vevo F2 device driving multiple high frequency probes (from 22 MHz to 50 MHz). Propagation of the SW was then followed by ultrafast plane wave imaging with a framerate of 1800 Hz. Stiffness maps obtained with the VevoF2 system were compared with those obtained with the SWE method (Aixplorer, Supersonic Imagine). Results of the shear wave dispersion showed that the phantom act as a Voigt model. At high frequency (VevoF2 at 22 MHz), the mean velocities of the dispersion curve for the medium and the inclusion are 2.11 ± 0.31 m/s (2.90 m/s for the Aixplorer) and 4.42 ± 0.63 m/s (4.52 m/s for the Aixplorer) respectively. Similar results are obtained at higher frequencies up to 50 MHz using 4 different probes. Although we must carefully consider the results since the methods (SWE vs vibrator) used are different, the feasibility of performing transient elastography at very high US frequencies has been demonstrated, thus providing higher spatial resolution for small biological soft tissues investigation. The long-term side effects such as breast fibrosis affect 20 to 50 % of women receiving breast-cancer radiotherapy (RT). This study’s objective is to determine if acute breast toxicity measured at the end of RT can predict long-term (1 year) toxicity using ultrasound radiomics and machine learning.Sixty-nine patients receiving RT for breast cancer were enrolled in the longitudinal ultrasound study. Each patient received ultrasound scans on the last day of RT and 1-year post RT on both treated and untreated breasts at five locations (12, 3, 6 and 9 o’clock and tumor bed). From breast ultrasound images, 163 radiomic features including first-order, gray-level co-occurrence matrix, gray level run length matrixgray level size zone matrix, and multiple gray level size zone matrix features, were extracted from the region of interest (3.5 cm width and 0.6 cm depth from the skin sur-face). Breast toxicity was scored in 4 levels (none, mild, moderate and severe) by two experienced ultrasound experts. After multicollinearity check and feature selection, selected early-stage radiomic features were employed to predict the late toxicity using five common machine learning classifiers: K-Nearest Neighbors (KNN), averaged Neural Networks (avNNet), random forest (RF), eXtreme Gradient Boosting (XGBoost), and Support Vector Machine (SVM). Repeated five-fold cross validation was used to evaluate the model performance. For binary classification of with and without late toxicity, the best predictive model is RF yielding an AUC (area under ROC curve) of 0.77 [CI: 0.7-0.84], sensitivity of 0.71 [CI: 0.61-0.8], and specificity of 0.68 [CI: 0.57-0.77]. For binary classification of no/mild and moderate/severe late toxicity, the RF model achieves an AUC of cancer-related The treatment includes followed Recent studies that Current imaging modalities monitoring can yield structural information but not mechanical properties of the tumor. Therefore, to measure the changes in the mechanical properties in response to We hypothesize that the responsiveness to radiotherapy is linked to changes in the tumor biomechanics. To test this we used shear wave elasticity imaging and frequency shift method to measure the temporal changes of shear wave speed (SWS) and attenuation in mice with 0.24 m/s for control, non-responsive and responsive tumors, respectively. The average attenuations were 3.18 ± 0.55Np/mm, 3.87 ± 0.71Np/mm, 5.18 ± 1.02Np/mm for control, non-responsive and responsive tumors, respectively. The non-responsive group exhibited higher shear wave speed and lower attenuation compared to the responsive group. In conclusion, shear wave speed and attenuation vary between responsive and non-responsive tumors. Targeted prostate biopsies require precise needle insertion for proper lesion sampling. Most clinical biopsies are performed using hand-held needles and are subject to variability in needle deflection resulting in inaccurate locations of tissue sampling. Previous work has shown that acoustic radiation force impulse (ARFI) imaging can identify the majority of clinically significant cancer lesions > 0.5mL. (1) By approximating the lesion volume as a sphere, ARFI techniques can identify lesions with a diameter of > ~9.85mm, thereby requiring < 1cm of certainty in biopsy needle targeting. However, needle accuracy is limited during transperineal biopsy due to deflection as a result of the single-beveled needle tip geometry, needle deflection upon initial insertion, biopsy grid motion, and limitations in biopsy grid geometry. Similarly, the degree of deflection due to tip geometry varies with in and between By accuracy within range of identifiable