Quantitative Ultrasound Metrics Research Articles

Comparative Study of Raw Ultrasound Data Representations in Deep Learning to Classify Hepatic Steatosis

Adiposity accumulation in the liver is an early-stage indicator of non-alcoholic fatty liver disease. Analysis of ultrasound (US) backscatter echoes from liver parenchyma with deep learning (DL) may offer an affordable alternative for hepatic steatosis staging. The aim of this work was to compare DL classification scores for liver steatosis using different data representations constructed from raw US data. Steatosis in N=31 patients with confirmed or suspected non-alcoholic fatty liver disease was stratified based on fat-fraction cutoff values using magnetic resonance imaging as a reference standard. US radiofrequency (RF) frames (raw data) and clinical B-mode images were acquired. Intermediate image formation stages were modeled from RF data. Power spectrum representations and phase representations were also calculated. Co-registered patches were used to independently train 1-, 2- and 3-D convolutional neural networks (CNNs), and classifications scores were compared with cross-validation. There were 67,800 patches available for 2-D/3-D classification and 1,830,600 patches for 1-D classification. The results were also compared with radiologist B-mode annotations and quantitative ultrasound (QUS) metrics. Patch classification scores (area under the receiver operating characteristic curve [AUROC]) revealed significant reductions along successive stages of the image formation process (p < 0.001). Patient AUROCs were 0.994 for RF data and 0.938 for clinical B-mode images. For all image formation stages, 2-D CNNs revealed higher patch and patient AUROCs than 1-D CNNs. CNNs trained with power spectrum representations converged faster than those trained with RF data. Phase information, which is usually discarded in the image formation process, provided a patient AUROC of 0.988. DL models trained with RF and power spectrum data (AUROC=0.998) provided higher scores than conventional QUS metrics and multiparametric combinations thereof (AUROC=0.986). Radiologist annotations indicated lower hepatic steatosis classification accuracies (Acc=0.914) with respect to magnetic resonance imaging proton density fat fraction that DL models (Acc=0.989). Access to raw ultrasound data combined with artificial intelligence techniques may offer superior opportunities for quantitative tissue diagnostics than conventional sonographic images.

Comparison of Quantitative Ultrasound Methods to Classify Dystrophic and Obese Models of Skeletal Muscle.

In this study, we compared multiple quantitative ultrasound metrics for the purpose of differentiating muscle in 20 healthy, 10 dystrophic and 10 obese mice. High-frequency ultrasound scans were acquired on dystrophic (D2-mdx), obese (db/db) and control mouse hindlimbs. A total of 248 image features were extracted from each scan, using brightness-mode statistics, Canny edge detection metrics, Haralick features, envelope statistics and radiofrequency statistics. Naïve Bayes and other classifiers were trained on single and pairs of features. The a parameter from the Homodyned K distribution at 40 MHz achieved the best univariate classification (accuracy=85.3%). Maximum classification accuracy of 97.7% was achieved using a logistic regression classifier on the feature pair of a2 (K distribution) at 30 MHz and brightness-mode variance at 40MHz. Dystrophic and obese mice have muscle with distinct acoustic properties and can be classified to a high level of accuracy using a combination of multiple features.

Time-domain formulation for coherent plane wave compounding in horizontally layered media

The locations of origin of ultrasonic echoes from soft tissue in conventional ultrasound beamforming are based on measured time of flight. For simplicity it is usually assumed that the region of interest is homogeneous with a constant sound speed, but in many in vivo and benchtop applications the imaging region is composed of horizontal layers, each with a distinct sound speed. In addition to time-of-flight complications, sound speed differences between the layers result in refraction through the interface of both the transmitted pulse and those reflected from the tissue, further influencing the fidelity of beamforming for target localization or other quantitative ultrasound metrics. A formulation is presented for delay-and-sum coherent Plane Wave Compounding (PWC) that accounts for variation in sound speed and refraction of ultrasound pulses in a region with two horizontal layers. The algorithm is formulated completely in the time domain, enabling simple implementation or extension of existing homogeneous PWC algorithms. Simulated and experimental images are investigated for a typical two-layer benchtop configuration, where the sound speeds in the upper and lower layers are relatively slow (water or gel) and fast (muscle), respectively. Extension of the formulation to a medium with N distinct layers is discussed.

Association of Clinical Assessments of Hand Function and Quantitative Ultrasound Metrics in First Carpometacarpal Osteoarthritis.

Thumb carpometacarpal (CMC) osteoarthritis (OA), a degenerative condition affecting hand use, is typically evaluated through radiographs and clinical examination. Although this can determine treatment, it is difficult to evaluate functional limitations. Shear wave elastography (SWE) is a quantitative ultrasound technique that characterizes tissue stiffness. This pilot study aimed to establish data of the SWE findings in the thenar eminence muscles in patients with first CMC OA and correlate these findings with the clinical tests of hand function. This cross-sectional study correlated the SWE stiffness of thenar eminence muscles to clinical tests of hand function in patients with first CMC OA and in asymptomatic control subjects, using Spearman's correlation coefficient. Mean SWE values of the thenar eminence muscles in patients were compared with those in control subjects. The study was performed in a non-profit tertiary care hospital setting. Patients and control subjects were recruited on a volunteer basis. SWE values in the abductor pollicis brevis and flexor pollicis brevis muscles showed moderate to very strong correlation with multiple measures of hand function. Mean SWE values of the thenar eminence muscles in first CMC OA patients were lower than those in asymptomatic control subjects. Correlations between mean SWE values in the thenar eminence muscles and clinical measures of hand function suggest decreased function in subjects with less stiff thenar eminence muscles.

Effects of Home Exercises on Shoulder Pain and Pathology in Chronic Spinal Cord Injury: A Randomized Controlled Trial.

The aims of this study were to replicate findings that a home exercise program intervention improved shoulder pain and function and to assess changes in supraspinatus tendon pathology associated with the home exercise program in persons with chronic spinal cord injury. The study is a single-blind randomized controlled trial. Individuals with spinal cord injury of at least 1 yr and chronic shoulder pain of moderate or greater average intensity were enrolled. Participants were randomized to a 12-wk home exercise program consisting of strengthening and stretching exercises or to an education-only control group, with immediate postintervention and 4-wk postintervention (16 wks) follow-ups. The main outcome measures were self-report measures of shoulder pain and impairment, the Physical Examination of the Shoulder Scale, quantitative ultrasound metrics of the supraspinatus tendon, and the Ultrasound Shoulder Pathology Rating Scale. Thirty-two participants were randomized to home exercise program or education-only control condition. The mean ± SD age was 44.8 ± 12.5; 81.3% were male; 65.6% had motor complete paraplegia. Using a per-protocol, within-group analysis method, significant differences were observed between baseline and postintervention for the home exercise program group for the least pain intensity (P = 0.02), number of days with shoulder pain (P = 0.042), Physical Examination of the Shoulder Scale scores (dominant side, P = 0.036; nondominant side, P = 0.008), the Disabilities of the Arm, Shoulder, and Hand (P = 0.028), and the Patient Global Impression of Change (P = 0.015). The education-only control condition group demonstrated significant changes in average unpleasantness of shoulder pain after the intervention period (P = 0.049). Comparisons in changes from baseline between groups showed that the home exercise program group had greater improvements in nondominant-side Physical Examination of the Shoulder Scale scores and global impression of change, whereas the education-only control condition group had greater improvements in depressive symptoms.For quantitative ultrasound measures, no significant changes were found with within-group analyses for the home exercise program group, although the education-only control condition group demonstrated a decrease in tendon width in the nondominant-side supraspinatus tendon (P = 0.036). Comparison of changes between groups suggests that the education-only control condition group had a greater increase in dominant shoulder supraspinatus tendon ultrasound contrast at the end of the study. Changes in several measures of shoulder pain and function occurred after the home exercise program intervention, although the magnitude of changes was only significantly greater than those of the education-only control condition group for two measures. Significant changes in supraspinatus pathology were not detected with quantitative ultrasound metrics.

ARFI and shear wave imaging of the prostate to delineate clinically-significant cancers

Current prostate cancer screening methods involve non-targeted needle biopsies and the detection of clinically-insignificant lesions that receive excessive treatments, exposing patients to unnecessary adverse side effects and placing a burden on our health care systems. There is a strong clinical need for improved prostate imaging methods that are sensitive and specific for clinically-significant prostate cancer lesions to guide needle biopsies, target focal treatments, and improve overall patient outcomes. We have developed a 3D, in vivo Acoustic Radiation Force Impulse (ARFI) imaging and Shear Wave Elasticity Imaging (SWEI) system using a side-fire endorectal linear array with a custom rotation stage to collect prostate imaging volumes to identify regions of suspicion that should be targeted during biopsy procedures. We have validated that ARFI imaging can reliability identify regions of clinically-significant prostate cancer by comparing our imaging volumes with whole-mount histology in patients undergoing radical prostatectomy; however, there was opportunity to improve the diagnostic sensitivity and specificity by combining ARFI imaging data with SWEI. While ARFI images contain regions of relative contrast (decreased displacement) to delineate regions of suspicion, we hypothesized that SWEI could provide absolute shear wave speeds that could quantify the degree of suspicion of an identified lesion, with more aggressive lesions being stiffer and having higher shear wave speeds. In an ongoing clinical study of over 30 patients, we have demonstrated that the combination of ARFI and SWEI data can improve the diagnostic accuracy of clinically-significant lesion identification, while the inclusion of B-mode echogenicity was not found to improve diagnostic accuracy. More complex classification models to identify clinically-significant prostate cancer using ARFI and SWEI metrics are being developed using Linear Discriminant Analysis (LDA) and Support-Vector Machines (SVMs). These classification models will be further extended to explore the diagnostic value of additional quantitative ultrasound metrics derived from the raw radiofrequency (RF) data and other imaging methods commonly used in the diagnostic workup of prostate cancer patients, such as multiparametric magnetic resonance imaging (mp-MRI).

Increased Reliability of Quantitative Ultrasound Measures of the Supraspinatus Tendon Using Multiple Image Analysts and Analysis Runs

Quantitative ultrasound (QUS) is an inexpensive and promising tool for sensitive measurement of tendon pathology. However, few studies have reported the psychometric properties of measurements obtained using this technique for assessments of the supraspinatus tendon. The present study was undertaken to determine the variance contributed by several sources of error (participant, ultrasound operator, image analyst, analysis session) to QUS measures of the supraspinatus tendon. Transverse images of the supraspinatus tendon were captured from eleven subjects (22 shoulders) by two ultrasonographers, and each image was analyzed by two image analysts who each completed two analysis runs. Generalizability theory and intraclass correlations were used to assess the reliability of seven QUS metrics. Measures of tendon/cartilage thickness demonstrated the greatest degree of overall dependability (ϕ = 0.84), followed by echogenicity (ϕ = 0.56), variance (ϕ = 0.55), and entropy (ϕ = 0.47), suggesting that these measures of the supraspinatus tendon may be promising metrics for assessing differences in tendon health. Interrater reliability between ultrasound operators ranged from low to moderate for different QUS metrics, but using more than one image analyst and performing repeated measurement analysis runs on each image help increase reliability of QUS measures for the supraspinatus tendon.