Diagnostic Accuracy of Echo Envelope Statistical Modeling Compared to B-Mode and Power Doppler Ultrasound Imaging in Patients With Clinically Diagnosed Lateral Epicondylosis of the Elbow

Assessing the treatment response of lateral elbow tendinopathy using time-dependent ultrasonography, Doppler imaging, and elastography

OBJECTIVE

To investigate the structural alterations, neovascularity, and elasticity of tendons and the relationship between elasticity and the Patient Rated Tennis Elbow Evaluation score after undergoing US-guided fenestration or surgery in patients with chronic lateral elbow tendinopathy.

METHODS

Participants from the per-protocol population of a randomized trial conducted between October 2016 and June 2020 were included. The surgery and fenestration groups included 24 (mean age, 50 ± 7 years [standard deviation], 10 men) and 29 (47 ± 8 years, 18 men) participants, respectively. Ultrasound exams were performed at baseline, 6 months, and 12 months. Statistical analyses included linear mixed effects and generalized equation estimation models.

RESULTS

Fenestration had no significant impact on tendon thickness (p = 0.46). Conversely, surgery significantly increased tendon thickness at 6 months (p < 0.0001) and remained elevated at 12 months (p = 0.04). Tendon echostructure exhibited a group effect (p = 0.03), indicating a higher proportion of pathological scores in the surgery group post-intervention compared to the fenestration group. Both groups showed a similar reduction in neovascularity from 6 to 12 months postintervention (p = 0.006). Shear-wave velocity increased in the fenestration group at 6 months (p = 0.04), while the surgery group experienced a nonsignificant decrease at 6 months, with some improvement at 12 months (p = 0.08). Changes in shear-wave velocity did not correlate with clinical outcome.

CONCLUSIONS

Fenestration and surgery reduced tendon neovascularity over time. Unlike surgery, fenestration did not impact tendon size while improving tendon echostructure and elasticity.

CRITICAL RELEVANCE STATEMENT

Fenestration and surgery equally alleviated symptoms and decreased tendon neovascularity in lateral elbow tendinopathy; however, fenestration did not alter tendon thickness and improved echostructure and shear-wave velocity, suggesting shear-wave velocity's potential for quantitatively monitoring tendon elasticity during healing.

KEY POINTS

Reliable markers for monitoring healing response and informing treatment protocols in elbow tendinopathy are lacking. Fenestration and surgery reduced tendon neovascularity, while fenestration improved tendon echostructure and shear-wave velocity. Shear-wave velocity may provide quantitative measures to monitor tendon elasticity in response to treatment.

Parallelized ultrasound homodyned-K imaging based on a generalized artificial neural network estimator

The homodyned-K (HK) distribution model is a generalized backscatter envelope statistical model for ultrasound tissue characterization, whose parameters are of physical meaning. To estimate the HK parameters is an inverse problem, and is quite complicated. Previously, we proposed an artificial neural network (ANN) estimator and an improved ANN (iANN) estimator for estimating the HK parameters, which are fast and flexible. However, a drawback of the conventional ANN and iANN estimators consists in that they use Monte Carlo simulations under known values of HK parameters to generate training samples, and thus the ANN and iANN models have to be re-trained when the size of the test sets (or of the envelope samples to be estimated) varies. In addition, conventional ultrasound HK imaging uses a sliding window technique, which is non-vectorized and does not support parallel computation, so HK image resolution is usually sacrificed to ensure a reasonable computation cost. To this end, we proposed a generalized ANN (gANN) estimator in this paper, which took the theoretical derivations of feature vectors for network training, and thus it is independent from the size of the test sets. Further, we proposed a parallelized HK imaging method that is based on the gANN estimator, which used a block-based parallel computation method, rather than the conventional sliding window technique. The gANN-based parallelized HK imaging method allowed a higher image resolution and a faster computation at the same time. Computer simulation experiments showed that the gANN estimator was generally comparable to the conventional ANN estimator in terms of HK parameter estimation performance. Clinical experiments of hepatic steatosis showed that the gANN-based parallelized HK imaging could be used to visually and quantitatively characterize hepatic steatosis, with similar performance to the conventional ANN-based HK imaging that used the sliding window technique, but the gANN-based parallelized HK imaging was over 3 times faster than the conventional ANN-based HK imaging. The parallelized computation method presented in this work can be easily extended to other quantitative ultrasound imaging applications.

Liver Fibrosis Assessment Using Radiomics of Ultrasound Homodyned-K imaging Based on the Artificial Neural Network Estimator

The homodyned-K distribution is an important ultrasound backscatter envelope statistics model of physical meaning, and the parametric imaging of the model parameters has been explored for quantitative ultrasound tissue characterization. In this paper, we proposed a new method for liver fibrosis characterization by using radiomics of ultrasound backscatter homodyned-K imaging based on an improved artificial neural network (iANN) estimator. The iANN estimator was used to estimate the ultrasound homodyned-K distribution parameters k and α from the backscattered radiofrequency (RF) signals of clinical liver fibrosis (n = 237), collected with a 3-MHz convex array transducer. The RF data were divided into two groups: Group I corresponded to liver fibrosis with no hepatic steatosis (n = 94), and Group II corresponded to liver fibrosis with mild to severe hepatic steatosis (n = 143). The estimated homodyned-K parameter values were then used to construct k and α parametric images using the sliding window technique. Radiomics features of k and α parametric images were extracted, and feature selection was conducted. Logistic regression classification models based on the selected radiomics features were built for staging liver fibrosis. Experimental results showed that the proposed method is overall superior to the radiomics method of uncompressed envelope images when assessing liver fibrosis. Regardless of hepatic steatosis, the proposed method achieved the best performance in staging liver fibrosis ≥F1, ≥F4, and the area under the receiver operating characteristic curve was 0.88, 0.85 (Group I), and 0.85, 0.86 (Group II), respectively. Radiomics has improved the ability of ultrasound backscatter statistical parametric imaging to assess liver fibrosis, and is expected to become a new quantitative ultrasound method for liver fibrosis characterization.

Ultrasound Scatterer Density Classification Using Convolutional Neural Networks and Patch Statistics

Quantitative ultrasound (QUS) can reveal crucial information on tissue properties, such as scatterer density. If the scatterer density per resolution cell is above or below 10, the tissue is considered as fully developed speckle (FDS) or underdeveloped speckle (UDS), respectively. Conventionally, the scatterer density has been classified using estimated statistical parameters of the amplitude of backscattered echoes. However, if the patch size is small, the estimation is not accurate. These parameters are also highly dependent on imaging settings. In this article, we adapt convolutional neural network (CNN) architectures for QUS and train them using simulation data. We further improve the network's performance by utilizing patch statistics as additional input channels. Inspired by deep supervision and multitask learning, we propose a second method to exploit patch statistics. We evaluate the networks using simulation data and experimental phantoms. We also compare our proposed methods with different classic and deep learning models and demonstrate their superior performance in the classification of tissues with different scatterer density values. The results also show that we are able to classify scatterer density in different imaging parameters with no need for a reference phantom. This work demonstrates the potential of CNNs in classifying scatterer density in ultrasound images.

Parameter estimation of the homodyned K distribution based on an artificial neural network for ultrasound tissue characterization

The homodyned K (HK) distribution allows a general description of ultrasound backscatter envelope statistics with specific physical meanings. In this study, we proposed a new artificial neural network (ANN) based parameter estimation method of the HK distribution. The proposed ANN estimator took advantages of ANNs in learning and function approximation and inherited the strengths of conventional estimators through extracting five feature parameters from backscatter envelope signals as the input of the ANN: the signal-to-noise ratio (SNR), skewness, kurtosis, as well as X- and U-statistics. Computer simulations and clinical data of hepatic steatosis were used for validations of the proposed ANN estimator. The ANN estimator was compared with the RSK (the level-curve method that uses SNR, skewness, and kurtosis based on the fractional moments of the envelope) and XU (the estimation method based on X- and U-statistics) estimators. Computer simulation results showed that the relative bias was best for the XU estimator, whilst the normalized standard deviation was overall best for the ANN estimator. The ANN estimator was almost one order of magnitude faster than the RSK and XU estimators. The ANN estimator also yielded comparable diagnostic performance to state-of-the-art HK estimators in the assessment of hepatic steatosis. The proposed ANN estimator has great potential in ultrasound tissue characterization based on the HK distribution.

Does Artificial Intelligence Outperform Natural Intelligence in Interpreting Musculoskeletal Radiological Studies? A Systematic Review

BACKGROUND

Machine learning (ML) is a subdomain of artificial intelligence that enables computers to abstract patterns from data without explicit programming. A myriad of impactful ML applications already exists in orthopaedics ranging from predicting infections after surgery to diagnostic imaging. However, no systematic reviews that we know of have compared, in particular, the performance of ML models with that of clinicians in musculoskeletal imaging to provide an up-to-date summary regarding the extent of applying ML to imaging diagnoses. By doing so, this review delves into where current ML developments stand in aiding orthopaedists in assessing musculoskeletal images.

QUESTIONS/PURPOSES

This systematic review aimed (1) to compare performance of ML models versus clinicians in detecting, differentiating, or classifying orthopaedic abnormalities on imaging by (A) accuracy, sensitivity, and specificity, (B) input features (for example, plain radiographs, MRI scans, ultrasound), (C) clinician specialties, and (2) to compare the performance of clinician-aided versus unaided ML models.

METHODS

A systematic review was performed in PubMed, Embase, and the Cochrane Library for studies published up to October 1, 2019, using synonyms for machine learning and all potential orthopaedic specialties. We included all studies that compared ML models head-to-head against clinicians in the binary detection of abnormalities in musculoskeletal images. After screening 6531 studies, we ultimately included 12 studies. We conducted quality assessment using the Methodological Index for Non-randomized Studies (MINORS) checklist. All 12 studies were of comparable quality, and they all clearly included six of the eight critical appraisal items (study aim, input feature, ground truth, ML versus human comparison, performance metric, and ML model description). This justified summarizing the findings in a quantitative form by calculating the median absolute improvement of the ML models compared with clinicians for the following metrics of performance: accuracy, sensitivity, and specificity.

RESULTS

ML models provided, in aggregate, only very slight improvements in diagnostic accuracy and sensitivity compared with clinicians working alone and were on par in specificity (3% (interquartile range [IQR] -2.0% to 7.5%), 0.06% (IQR -0.03 to 0.14), and 0.00 (IQR -0.048 to 0.048), respectively). Inputs used by the ML models were plain radiographs (n = 8), MRI scans (n = 3), and ultrasound examinations (n = 1). Overall, ML models outperformed clinicians more when interpreting plain radiographs than when interpreting MRIs (17 of 34 and 3 of 16 performance comparisons, respectively). Orthopaedists and radiologists performed similarly to ML models, while ML models mostly outperformed other clinicians (outperformance in 7 of 19, 7 of 23, and 6 of 10 performance comparisons, respectively). Two studies evaluated the performance of clinicians aided and unaided by ML models; both demonstrated considerable improvements in ML-aided clinician performance by reporting a 47% decrease of misinterpretation rate (95% confidence interval [CI] 37 to 54; p < 0.001) and a mean increase in specificity of 0.048 (95% CI 0.029 to 0.068; p < 0.001) in detecting abnormalities on musculoskeletal images.

CONCLUSIONS

At present, ML models have comparable performance to clinicians in assessing musculoskeletal images. ML models may enhance the performance of clinicians as a technical supplement rather than as a replacement for clinical intelligence. Future ML-related studies should emphasize how ML models can complement clinicians, instead of determining the overall superiority of one versus the other. This can be accomplished by improving transparent reporting, diminishing bias, determining the feasibility of implantation in the clinical setting, and appropriately tempering conclusions.

LEVEL OF EVIDENCE

Level III, diagnostic study.

Value of homodyned K distribution in ultrasound parametric imaging of hepatic steatosis: An animal study

Ultrasound is the first-line tool for screening hepatic steatosis. Statistical distributions can be used to model the backscattered signals for liver characterization. The Nakagami distribution is the most frequently adopted model; however, the homodyned K (HK) distribution has received attention due to its link to physical meaning and improved parameter estimation through X- and U-statistics (termed "XU"). To assess hepatic steatosis, we proposed HK parametric imaging based on the α parameter (a measure of the number of scatterers per resolution cell) calculated using the XU estimator. Using a commercial system equipped with a 7-MHz linear array transducer, phantom experiments were performed to suggest an appropriate window size for α imaging using the sliding window technique, which was further applied to measuring the livers of rats (n = 66) with hepatic steatosis induced by feeding the rats a methionine- and choline-deficient diet. The relationships between the α parameter, the stage of hepatic steatosis, and histological features were verified by the correlation coefficient r, one-way analysis of variance, and regression analysis. The phantom results showed that the window side length corresponding to five times the pulse length supported a reliable α imaging. The α parameter showed a promising performance for grading hepatic steatosis (p < 0.05; r² = 0.68). Compared with conventional Nakagami imaging, α parametric imaging provided significant information associated with fat droplet size (p < 0.05; r² = 0.53), enabling further analysis and evaluation of severe hepatic steatosis.

Hepatic steatosis assessment using ultrasound homodyned-K parametric imaging: the effects of estimators

BACKGROUND

The homodyned-K (HK) distribution is an important statistical model for describing ultrasound backscatter envelope statistics. HK parametric imaging has shown potential for characterizing hepatic steatosis. However, the feasibility of HK parametric imaging in assessing human hepatic steatosis in vivo remains unclear.

METHODS

In this paper, ultrasound HK μ parametric imaging was proposed for assessing human hepatic steatosis in vivo. Two recent estimators for the HK model, RSK (the level-curve method that uses the signal-to-noise ratio (SNR), skewness, and kurtosis based on the fractional moments of the envelope) and XU (the estimation method based on the first moment of the intensity and two log-moments, namely X- and U-statistics), were investigated. Liver donors (n=72) and patients (n=204) were recruited to evaluate hepatic fat fractions (HFFs) using magnetic resonance spectroscopy and to evaluate the stages of fatty liver disease (normal, mild, moderate, and severe) using liver biopsy with histopathology. Livers were scanned using a 3-MHz ultrasound to construct μ _RSK and μ _XU images to correlate with HFF analyses and fatty liver stages. The μ _RSK and μ _XU parametric images were constructed using the sliding window technique with the window side length (WSL) =1-9 pulse lengths (PLs). The diagnostic values of the μ _RSK and μ _XU parametric imaging methods were evaluated using receiver operating characteristic (ROC) curves.

RESULTS

For the 72 participants in Group A, the μ _RSK parametric imaging with WSL =2-9 PLs exhibited similar correlation with log₁₀(HFF), and the μ _RSK parametric imaging with WSL = 3 PLs had the highest correlation with log₁₀(HFF) (r=0.592); the μ _XU parametric imaging with WSL =1-9 PLs exhibited similar correlation with log₁₀(HFF), and the μ _XU parametric imaging with WSL =1 PL had the highest correlation with log₁₀(HFF) (r=0.628). For the 204 patients in Group B, the areas under the ROC (AUROCs) obtained using μ _RSK for fatty stages ≥ mild (AUROC₁), ≥ moderate (AUROC₂), and ≥ severe (AUROC₃) were (AUROC₁, AUROC₂, AUROC₃) = (0.56, 0.57, 0.53), (0.68, 0.72, 0.75), (0.73, 0.78, 0.80), (0.74, 0.77, 0.79), (0.74, 0.78, 0.79), (0.75, 0.80, 0.82), (0.74, 0.77, 0.83), (0.74, 0.78, 0.84) and (0.73, 0.76, 0.83) for WSL =1, 2, 3, 4, 5, 6, 7, 8 and 9 PLs, respectively. The AUROCs obtained using μ _XU for fatty stages ≥ mild, ≥ moderate, and ≥ severe were (AUROC₁, AUROC₂, AUROC₃) = (0.75, 0.83, 0.81), (0.74, 0.80, 0.80), (0.76, 0.82, 0.82), (0.74, 0.80, 0.84), (0.76, 0.80, 0.83), (0.75, 0.80, 0.84), (0.75, 0.79, 0.85), (0.75, 0.80, 0.85) and (0.73, 0.77, 0.83) for WSL = 1, 2, 3, 4, 5, 6, 7, 8 and 9 PLs, respectively.

CONCLUSIONS

Both the μ _RSK and μ _XU parametric images are feasible for evaluating human hepatic steatosis. The WSL exhibits little impact on the diagnosing performance of the μ _RSK and μ _XU parametric imaging. The μ _XU parametric imaging provided improved performance compared to the μ _RSK parametric imaging in characterizing human hepatic steatosis in vivo.