Caterina M. Gallippi

Purpose Hybrid exoskeletons for spinal cord injury rehabilitation combine functional electrical stimulation (FES) with robotic assistance, but muscle fatigue limits FES effectiveness. Improved fatigue measurements could improve exoskeleton control by optimizing assistance. This study evaluates Viscoelastic Response (VisR) ultrasound imaging as a method for monitoring muscle fatigue through changes in muscle stiffness. Methods VisR ultrasound was used to assess the rectus femoris in five unimpaired participants and three with spinal cord injury (SCI). Changes in relative elasticity (RE) from baseline were measured during FES-induced and volitional contractions, as well as in post-contraction stiffness following FES stimulation. Both transverse and longitudinal transducer orientations were used for data collection to evaluate the impact of transducer orientation. Results In control participants, percent changes in RE were strongly correlated with FES-induced torque, with no significant effect of transducer orientation. A similar relationship was observed in SCI participants when the transducer was oriented transversely but not longitudinally. In both control and SCI participants, percent changes in RE during volitional contraction were lower than those predicted by FES-based linear models. Post-contraction RE values did not differ significantly from pre-contraction values after FES. Conclusion VisR ultrasound effectively tracks muscle stiffness during isometric contractions, with percent changes in RE closely reflecting FES-induced torque output. Findings suggest VisR may provide relevant feedback on actively contracting muscles, supporting its potential as a tool for monitoring fatigue and muscle activation to inform adaptive control strategies in rehabilitative hybrid exoskeletons.

We herein propose Deep-DoPIo, a deep learning–based framework that predicts shear elastic modulus from paired narrow- and wide-track displacement profiles generated by a single acoustic radiation force (ARF) push. Unlike the traditional Double Profile Intersection (DoPIo) ultrasound technique, which relates the intersection time (t_int) between these profiles to shear modulus through a linear empirical model, Deep-DoPIo leverages neural networks to learn this mapping directly from the displacement data. Four architectures were evaluated: a feed-forward neural network (FFNN), a one-dimensional convolutional neural network (CNN), an encoder–transformer (E-Tr), and a one-dimensional vision transformer (ViT). Models were trained on simulated displacement data and fine-tuned on calibrated phantom measurements for domain adaptation. Across simulated materials, traditional DoPIo yielded increasing mean absolute errors (MAE) with increasing material stiffness, with MAE exceeding 13 kPa for 28 kPa material. Conversely, all examined DNN models maintained MAEs below 0.5 kPa regardless of material stiffness. In calibrated phantoms, FFNN and CNN produced similar error trends to traditional DoPIo (MAE ≈ 4-8 kPa in a 22.20 kPa medium), indicating limited benefit over the empirical approach. By contrast, transformer-based models (E-Tr and ViT) consistently yielded low errors (<0.2 kPa) across all media of varying stiffness. In a custom liver phantom with a gelatin–graphite inclusion, traditional DoPIo substantially underestimated elasticity relative to shear wave elasticity imaging (SWEI), with errors approaching 50%. In contrast, transformer-based elasticity predictions closely matched those of SWEI, differing by less than 4% in the inclusion and less than 9% in the background. These results demonstrate that transformer-based architectures can generalize from simulation to both calibrated and excised tissue phantoms, providing accurate shear modulus estimates under conditions where traditional DoPIo struggles. Future work will evaluate performance in vivo and systematically compare Deep-DoPIo against SWEI under clinically relevant conditions.

Viscoelastic response (VisR) ultrasound has been developed by our group to interrogate tissue stiffness and viscosity. VisR has several potential advantages for breast cancer diagnostic imaging such being non-invasive and low-cost. Because ultrasound can penetrate dense breasts more effectively than mammograms, it may improve the detection of malignant masses in women with dense breasts. VisR-based estimates of stiffness, viscosity, and anisotropy have been shown in our preliminary studies to discriminate malignant and benign breast lesions. However, a potential limitation of VisR could be dependence on tissue pre-loading from applied surface compression by the practitioner. We conducted an IRB-approved clinical study of 20 women with no known breast pathologies to assess the impact of compression on VisR measurements of peak displacement (PD), relative elasticity (RE), relative viscosity (RV), and degree of anisotropy (DoA). Participants were between the ages of 30-90, and 10/20 had mammographically dense breasts. We found that surface compression significantly affected measurements of PD, RE, and RV in breast tissue, in vivo. In particular, in women with dense breasts, stiffness (via PD and RE) increased significantly with applied compression. DoA of PD, RE, and RV increased, decreased, or stayed the same with compression. No significant difference was found in DoA with compression between the breast density groups. Based on these findings, we recommend that surface compression be standardized and monitored when using VisR for clinical breast imaging, especially in women with dense breasts. Further studies are needed to identify an optimal strain range for VisR measurement repeatability.

This study evaluates the potential for interrogating the Young’s elastic moduli in anisotropic media, including tissue, using Double Profile Intersection (DoPIo) ultrasound. DoPIo is an on-axis acoustic radiation force (ARF)-based elasticity imaging method that quantifies shear elasticity without relying on shear wave propagation. It is hypothesized that by applying a range of ARF excitations that are not perpendicular to the axis of symmetry (AoS) of transversely isotropic (TI) materials and monitoring the resultant variation in DoPIo-measured elasticity versus excitation angle, the Young’s elastic modulus may be interrogated in addition to the shear elastic modulus. The hypothesis was tested in silico, and results suggested that while DoPIo outcomes measured at normal (90°) ARF-AoS incidence were related to the shear elastic modulus alone, variation in DoPIo-derived elasticity over ARF-AoS incidence angle (defined as $\Delta \textit {Elasticity}$ ) exhibited a strong linear correlation with the longitudinal Young’s modulus ( ${E}_{L}$ ). The results suggest that ${E}_{L}$ evaluated by the rate of change of $\Delta \textit {Elasticity}$ with ARF-AoS incidence angle may serve as a novel biomarker for characterizing elastically anisotropic tissues such as kidney, skeletal muscle, and breast.

Carotid atherosclerosis is a major cause of ischemic stroke, and the ability to non-invasively assess plaque composition and structure is critical to effective stroke risk assessment. Carotid plaque components are delineated noninvasively by Acoustic Radiation Force Impulse (ARFI)-derived Variance of Acceleration, evaluated as its decadic log (log(VoA)). To date, this log(VoA) parameter has been calculated by isolating the variance in ARFI-induced displacement profiles using the second-order time derivative (SOTD), a high-pass filtering operation. The purpose of this study was to compare the performance of the SOTD filter to various other filtering methods in application to delineating human carotid plaque components, In Vivo. Specifically, the SOTD filter was compared to Principal Component Analysis (PCA), Finite Impulse Response (FIR), Infinite Impulse Response (IIR), and mean-center spatial (MCS) filters. Filter performances were evaluated in terms of the resulting log(VoA) generalized contrast-to-noise ratio (gCNR) for distinguishing plaque features in human carotid plaques, In Vivo, which were validated by spatially aligned histology. Results indicated that the SOTD filter consistently provided the highest gCNR for most plaque components, whereas the performances yielded by the other filters were more variable. The study demonstrated that the SOTD filter remains the preferred method for log(VoA) calculation due to its effectiveness for delineating carotid plaque features.

PurposeWe propose and evaluate multimodal deep learning (DL) approaches that combine ultrasound (US) B-mode and Nakagami parametric images for breast tumor classification. It is hypothesized that integrating tissue brightness information from B-mode images with scattering properties from Nakagami images will enhance diagnostic performance compared with single-input approaches.ApproachAn EfficientNetV2B0 network was used to develop multimodal DL frameworks that took as input (i) numerical two-dimensional (2D) maps or (ii) rendered red-green-blue (RGB) representations of both B-mode and Nakagami data. The diagnostic performance of these frameworks was compared with single-input counterparts using 831 US acquisitions from 264 patients. In addition, gradient-weighted class activation mapping was applied to evaluate diagnostically relevant information utilized by the different networks.ResultsThe multimodal architectures demonstrated significantly higher area under the receiver operating characteristic curve (AUC) values (p<0.05) than their monomodal counterparts, achieving an average improvement of 10.75%. In addition, the multimodal networks incorporated, on average, 15.70% more diagnostically relevant tissue information. Among the multimodal models, those using RGB representations as input outperformed those that utilized 2D numerical data maps (p<0.05). The top-performing multimodal architecture achieved a mean AUC of 0.896 [95% confidence interval (CI): 0.813 to 0.959] when performance was assessed at the image level and 0.848 (95% CI: 0.755 to 0.903) when assessed at the lesion level.ConclusionsIncorporating B-mode and Nakagami information together in a multimodal DL framework improved classification outcomes and increased the amount of diagnostically relevant information accessed by networks, highlighting the potential for automating and standardizing US breast cancer diagnostics to enhance clinical outcomes.

Ultrasound elastography is increasingly being used alongside mammography for breast cancer diagnosis, particularly in women with radiographically dense breasts. The elastogram-to-B-Mode ratio (E/B), which compares lesion sizes in B-Mode and stiffness images, has been shown to differentiate malignant (E/B >1) from benign (E/B 1 in malignant lesions and <1 in benign lesions) to those observed in stiffness images. Additionally, E/B values calculated from RE alone, RV alone, or a combination of RE and RV achieved 4-16% higher AUCs for discriminating malignant lesions compared to E/B derived solely from Acoustic Radiation Force Impulse (ARFI) peak displacement. These results suggest that incorporating VisR-derived viscosity and elasticity metrics into E/B calculations could significantly improve diagnostic accuracy for breast cancer detection.

This study evaluates the potential for interrogating the Young's elastic moduli in anisotropic media, including tissue, using Viscoelastic Response (VisR) ultrasound. VisR is an on-axis acoustic radiation force (ARF)-based elasticity imaging method that has been demonstrated previously for assessing the shear elastic moduli of transversely isotropic (TI) materials when the applied ARF excitation was incident normal to the axis of symmetry (AoS). It is hypothesized that by applying a range of non-normal ARF excitations and monitoring the percent change in VisR-derived relative elasticity (RE) versus ARF-AoS incidence angle, both the shear and the Young's elastic moduli may be interrogated. The hypothesis was tested using in silico experiments, which showed that while RE measured at normal ARF-AoS incidence was related to only longitudinal shear modulus (μ _L ) alone, the percent change in RE over ARF-AoS incidence angle (defined as Δ RE ) exhibited a strong linear correlation with the ratio of longitudinal shear-to-Young’s moduli (μ _L / E_L ), with correlation coefficients of 0.97-0.99. Additionally, the linear regression slopes of Δ RE versus ARF-AoS incidence angle statistically discriminated TI materials with μ _L / E_L ratios that were as little as 7% different. These findings were validated ex vivo in chicken pectoralis major and bovine longissimus dorsi muscles, where the rate of change in of Δ RE versus ARF-AoS incidence angle distinguished the two muscles with statistical significance. The results obtained in this study suggest that the rate of change of Δ RE with ARF-AoS incidence angle may serve as a novel, semi-quantitative biomarker for characterizing anisotropic tissues such as kidney, skeletal muscle, and breast.

Objective: Afferent stimulation of peripheral nerves has emerged as a non-invasive approach to suppress wrist tremors in persons with Essential Tremor (ET) and Parkinson' s disease (PD). Electromyography (EMG), a standard sensing modality for tremor characterization during afferent stimulation, has a significant drawback. Due to stimulation artifacts, measuring tremor frequency with EMG and applying stimulation simultaneously is often tricky. This paper investigates a stimulation artifact-free approach that uses real-time ultrasound (US) imaging-derived signals to cue afferent stimulation. Methods: We derived a real-time US imaging-based tissue displacement metric to characterize the wrist tremor in participants with PD or ET. We also compared the metric' s feasibility in detecting the wrist tremor with conventional sensors. We further determined an effective afferent stimulation frequency for each participant that achieved the best tremor suppression. Finally, the metric was evaluated to trigger afferent stimulation (on or off) on the onset or subsidence of the wrist tremor in four participants with PD or ET (two PD, two ET). Results: Ultrasound imaging can effectively measure the tremor frequency with no significant difference from gold-standard sensors such as EMG and IMU. The US metric-triggered personalized stimulation frequency achieved tremor suppression ratios ranging from 20-73 % in two PD and two ET participants. Conclusion: These findings indicate that a stimulation artifact-free US-imaging-based metric can simultaneously measure individual tremor characteristics and trigger afferent stimulation. Significance: Our work lays the foundation for new US imaging-based, non-invasive afferent stimulation paradigms for tremor suppression that can potentially benefit more than 11 million people with ET and PD.

We present a machine learning method to directly estimate viscoelastic moduli from displacement time-series profiles generated by Viscoelastic Response (VisR) ultrasound excitations. VisR uses two colocalized acoustic radiation force pushes to approximate tissue viscoelastic creep response and tracks displacements on-axis to measure the material relaxation. A fully-connected neural network is trained to learn a nonlinear mapping from VisR displacements, the push focal depth, and the measurement axial depth to the material elastic and viscous moduli. In this work, we assess the validity of Quantitative Viscoelastic Response (QVisR) in simulated materials, propose a method of domain adaption to phantom VisR displacements, and show in vivo estimates from a clinically acquired dataset.