pith. sign in

arxiv: 1907.01649 · v1 · pith:3ULJAI5Xnew · submitted 2019-07-02 · 📡 eess.IV · cs.CV

Estimation of Absolute States of Human Skeletal Muscle via Standard B-Mode Ultrasound Imaging and Deep Convolutional Neural Networks

Ryan J. Cunningham , Ian D. Loram This is my paper

Pith reviewed 2026-05-25 10:16 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords ultrasound imagingdeep convolutional networksmuscle state estimationB-mode ultrasoundskeletal muscleelectromyographyjoint momentgeneralization
0
0 comments X

The pith

Deep convolutional neural networks predict absolute muscle activity, joint angle and moment from standard 2D ultrasound images in held-out individuals.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether 2D B-mode ultrasound images alone contain enough information for deep networks to recover the absolute activity-length-tension state of skeletal muscle. Synchronized ultrasound, EMG, joint angle and moment recordings were collected from the calf muscles of 32 participants while active contraction and passive stretch were varied independently. Segmented images of the medial gastrocnemius and soleus were used to train a convolutional network that outputs the three state variables. Cross-validation leaving out entire participants yielded mean accuracies of 55 percent for angle, 57 percent for EMG and 46 percent for moment. The result indicates that observation-only 2D ultrasound plus deep learning can supply a generalizable, drift-free readout of neural output, length and tension.

Core claim

With 2D US imaging, deep neural networks can encode in generalizable form the activity-length-tension state relationship of muscle. A convolutional network trained on segmented calf images predicts ankle joint angle, electromyography and joint moment to accuracies of 55±8 percent, 57±11 percent and 46±9 percent respectively across all 32 held-out participants under independent active and passive input variation, establishing that low-power 2D ultrasound can provide non-invasive estimation of neural output, length and tension.

What carries the argument

Deep convolutional neural network that maps segmented 2D ultrasound regions of interest directly to the three absolute state components (activity, joint angle, joint moment).

If this is right

  • Ultrasound could supply simultaneous, drift-free estimates of muscle activity, length and force where EMG and dynamometry are used separately today.
  • The activity-length-tension relationship could be measured non-invasively during natural movement.
  • The same imaging-plus-network approach could support diagnosis and monitoring in pain, injury, neurological conditions and ageing.
  • Real-time muscle-state feedback becomes feasible without implanted sensors or high-power imaging modalities.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same network architecture might generalize to other muscles if similar synchronized datasets are collected.
  • Accuracy could rise with larger training sets or 3D ultrasound volumes, moving the method closer to clinical thresholds.
  • Wearable ultrasound probes combined with the network could enable continuous muscle-state tracking outside the lab.
  • The encoding learned by the network may capture structural features that traditional image analysis misses.

Load-bearing premise

Data from 32 participants under controlled independent variation of active and passive inputs are sufficient to train a model that generalizes to completely new individuals.

What would settle it

Applying the trained network to a new cohort of participants and obtaining prediction accuracies indistinguishable from chance.

Figures

Figures reproduced from arXiv: 1907.01649 by Ian D. Loram, Ryan J. Cunningham.

Figure 1
Figure 1. Figure 1: Experimental setup. The participant stood upright on a foot pedal system (yellow), while strapped (red) at the chest to a backboard and observed an oscilloscope at eye level. An US probe (green) was attached to the left calf to image the gastrocnemius medialis (GM) and soleus (SO) muscles (right: grayscale). A wireless EMG sensor was attached to GM and to SO at standard locations (http://www.seniam.org/). … view at source ↗
read the original abstract

Objective: To test automated in vivo estimation of active and passive skeletal muscle states using ultrasonic imaging. Background: Current technology (electromyography, dynamometry, shear wave imaging) provides no general, non-invasive method for online estimation of skeletal intramuscular states. Ultrasound (US) allows non-invasive imaging of muscle, yet current computational approaches have never achieved simultaneous extraction nor generalisation of independently varying, active and passive states. We use deep learning to investigate the generalizable content of 2D US muscle images. Method: US data synchronized with electromyography of the calf muscles, with measures of joint moment/angle were recorded from 32 healthy participants (7 female, ages: 27.5, 19-65). We extracted a region of interest of medial gastrocnemius and soleus using our prior developed accurate segmentation algorithm. From the segmented images, a deep convolutional neural network was trained to predict three absolute, drift-free, components of the neurobiomechanical state (activity, joint angle, joint moment) during experimentally designed, simultaneous, independent variation of passive (joint angle) and active (electromyography) inputs. Results: For all 32 held-out participants (16-fold cross-validation) the ankle joint angle, electromyography, and joint moment were estimated to accuracy 55+-8%, 57+-11%, and 46+-9% respectively. Significance: With 2D US imaging, deep neural networks can encode in generalizable form, the activity-length-tension state relationship of muscle. Observation only, low power, 2D US imaging can provide a new category of technology for non-invasive estimation of neural output, length and tension in skeletal muscle. This proof of principle has value for personalised muscle diagnosis in pain, injury, neurological conditions, neuropathies, myopathies and ageing.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The manuscript reports a study using B-mode ultrasound images of the medial gastrocnemius and soleus from 32 participants. After applying a prior segmentation algorithm to extract ROIs, a CNN is trained to regress three absolute state variables (EMG activity, ankle joint angle, joint moment) under independent variation of active and passive inputs. In 16-fold cross-validation with held-out participants, the model achieves reported accuracies of 55±8% (angle), 57±11% (EMG), and 46±9% (moment). The central claim is that 2D US images contain generalizable information about the activity-length-tension relationship that can be extracted by deep networks.

Significance. If the reported generalization holds without leakage from the segmentation step and with a clearly defined performance metric, the work would demonstrate a non-invasive, drift-free estimator of intramuscular states from standard imaging. This could open a new category of technology for clinical monitoring in neuromuscular conditions. The use of held-out participants and synchronized multi-modal ground truth are positive design elements.

major comments (3)
  1. [Abstract] Abstract: the performance metric called 'accuracy' for continuous regression outputs (55±8%, 57±11%, 46±9%) is non-standard and its exact definition is never supplied. Without this, the numerical claims cannot be interpreted or compared to conventional regression metrics such as R², normalized MAE, or correlation coefficient.
  2. [Methods] Methods (ROI extraction paragraph): the pipeline first applies a prior segmentation algorithm to produce the input images for the CNN. Because joint angle is directly encoded in the geometric length and boundary positions of the segmented muscle, angle prediction can succeed from boundary geometry alone rather than from learned texture-to-state mappings. No ablation on raw (unsegmented) images, no held-out validation of the segmentation algorithm itself, and no quantitative check that segmentation accuracy does not leak length cues are provided.
  3. [Results] Results (cross-validation paragraph): the reported 55±8% accuracy on joint angle is only modestly above a trivial baseline for a continuous variable; combined with the segmentation step, this raises the possibility that the EMG and moment results (57±11%, 46±9%) partly reflect residual correlations rather than extraction of the claimed activity-length-tension relationship.
minor comments (1)
  1. [Abstract] Abstract: notation '55+-8%' should be standardized to '55 ± 8%'.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the detailed and constructive review. We address each major comment below with point-by-point responses. Where the manuscript requires clarification or additional analysis, we indicate planned revisions. We have been careful not to overstate what the original experiments demonstrate.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the performance metric called 'accuracy' for continuous regression outputs (55±8%, 57±11%, 46±9%) is non-standard and its exact definition is never supplied. Without this, the numerical claims cannot be interpreted or compared to conventional regression metrics such as R², normalized MAE, or correlation coefficient.

    Authors: We agree that the term 'accuracy' is non-standard for regression and that its precise definition was not supplied in the manuscript. In the revised version we will replace the reported figures with standard metrics (R², normalized MAE, and Pearson r) computed on the held-out participants and will update both the abstract and results sections accordingly. This change will allow direct comparison with conventional regression performance. revision: yes

  2. Referee: [Methods] Methods (ROI extraction paragraph): the pipeline first applies a prior segmentation algorithm to produce the input images for the CNN. Because joint angle is directly encoded in the geometric length and boundary positions of the segmented muscle, angle prediction can succeed from boundary geometry alone rather than from learned texture-to-state mappings. No ablation on raw (unsegmented) images, no held-out validation of the segmentation algorithm itself, and no quantitative check that segmentation accuracy does not leak length cues are provided.

    Authors: The segmentation step was introduced to restrict the network input to the muscle tissue of interest. We acknowledge that boundary geometry could contribute to angle prediction and that the absence of an ablation on raw images leaves open the possibility of leakage. In the revision we will add an explicit discussion of this limitation, including the fact that EMG and moment predictions are less directly supported by geometry. We cannot, however, supply new ablation results on raw images because those experiments were not performed. revision: partial

  3. Referee: [Results] Results (cross-validation paragraph): the reported 55±8% accuracy on joint angle is only modestly above a trivial baseline for a continuous variable; combined with the segmentation step, this raises the possibility that the EMG and moment results (57±11%, 46±9%) partly reflect residual correlations rather than extraction of the claimed activity-length-tension relationship.

    Authors: We accept that the angle result is modest and that residual correlations cannot be ruled out without further controls. In the revised manuscript we will add explicit baseline comparisons (mean predictor and linear regression on segmented length) and will discuss the extent to which the EMG and moment results can be attributed to learned textural features versus geometric cues. These additions will be placed in the results and discussion sections. revision: yes

standing simulated objections not resolved
  • Absence of ablation experiments on raw (unsegmented) ultrasound images and held-out validation of the segmentation algorithm, neither of which were conducted in the original study.

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes an empirical supervised learning pipeline: synchronized US, EMG, angle and moment data are collected from 32 participants under controlled variation of active and passive inputs; a prior segmentation algorithm extracts an ROI; a CNN is trained to regress the three state variables; performance is reported via 16-fold cross-validation on completely held-out participants. No mathematical derivation chain exists. The single self-reference to the authors' earlier segmentation work is a preprocessing detail whose validity is not required to establish the held-out prediction numbers. Those numbers are measured against independently recorded target variables on unseen subjects and therefore do not reduce to the training inputs by construction. No self-definitional loop, fitted-parameter-as-prediction, or load-bearing uniqueness theorem is present.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the CNN learning a mapping from segmented images to three continuous state variables; the dominant fitted elements are the network weights, and the approach inherits the accuracy of a prior segmentation algorithm.

free parameters (1)
  • CNN weights
    Network parameters are optimized on the ultrasound image dataset to regress the three target variables.
axioms (2)
  • domain assumption Prior segmentation algorithm accurately isolates medial gastrocnemius and soleus regions of interest
    All subsequent feature learning depends on the ROI extracted by the earlier segmentation method.
  • domain assumption Experimental protocol produces independent variation of passive (angle) and active (EMG) inputs across participants
    The training objective requires that the three targets can be disentangled from the images.

pith-pipeline@v0.9.0 · 5874 in / 1099 out tokens · 36328 ms · 2026-05-25T10:16:59.400888+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

84 extracted references · 84 canonical work pages

  1. [1]

    Isometric The pedal system was fixed at a neutral angle (flat feet), and participants observed an analogue oscilloscope. On the oscillo- scope, we displayed, side by side, a dot representing the ampli- tude of their filtered GM EMG signal, and a dot representing the amplitude of a fabricated (target) signal (see section III. B.). Participants were asked t...

    work page

  2. [2]

    On the oscil- loscope, we displayed, side by side, a dot representing the am- plitude o f their filtered GM EMG signal

    Passive Participants observed an analogue oscilloscope. On the oscil- loscope, we displayed, side by side, a dot representing the am- plitude o f their filtered GM EMG signal. Participants were asked to monitor and minimize any EMG activity by relaxing their muscles. The pedal system was driven using a fabri cated signal (c.f. III. B.). Participants were ...

    work page

  3. [3]

    Combined The pedal system was fixed at a neutral angle (flat feet), and participants observed an analogue oscilloscope. On the oscillo- scope, we displayed, side by side, a dot representing the ampli- tude of their filtered GM EMG signal, and a dot representing the amplitude of a fabricated (target) signal (III. B.). The pedal system was simultaneously dr...

    work page

  4. [4]

    (1) 5 Figure 4

    30𝜋 − 𝜋 2). (1) 5 Figure 4. Representative example of neural network performance. Each panel compares the neural network output (blue) with the labels (green) for a single participant over all 3 trial conditions (combined: left, isometric: middle, and passive: right). Units are mV, Nm and degrees for EMG, Moment and Joint Angle. Figure 5. Zoom portion of ...

    work page

  5. [5]

    we additionally address the deep muscle (SO), and this de- cision inspired a CNN architecture wherein a CNN model is created and applied per muscle, and the weights were shared between models as a regulariser (Fig. 3). We are not the first to use this type of architecture [71], though the application if it is novel. The learning objectives are different f...

    work page

  6. [6]

    The economic costs of pain in the United States,

    D. J. Gaskin and P. Richard, “The economic costs of pain in the United States,” J. Pain, vol. 13, no. 8, pp. 715–724, 2012

    work page 2012

  7. [7]

    Individualized exercise interventions for spinal pain,

    D. Falla and P. W. Hodges, “Individualized exercise interventions for spinal pain,” Exerc. Sport Sci. Rev., vol. 45, no. 2, pp. 105–115, 2017

    work page 2017

  8. [8]

    Muscle ultrasound analysis: Normal values and differentiation between myopathies and neuropathies,

    N. M. Maurits, A. E. Bollen, A. Windhausen, A. E. J. De Jager, and J. H. Van Der Hoeven, “Muscle ultrasound analysis: Normal values and differentiation between myopathies and neuropathies,” Ultrasound Med. Biol., vol. 29, no. 2, pp. 215–225, 2003

    work page 2003

  9. [9]

    Skeletal muscle ultrasound.,

    S. Pillen and N. van Alfen, “Skeletal muscle ultrasound.,” Neurol. Res., vol. 33, no. 10, pp. 1016– 24, 2011

    work page 2011

  10. [10]

    Ergonomic design and training for preventing work-related musculoskeletal disorders of the upper limb and neck in adults.,

    V. C. W. Hoe, D. M. Urquhart, H. L. Kelsall, and M. R. Sim, “Ergonomic design and training for preventing work-related musculoskeletal disorders of the upper limb and neck in adults.,” Cochrane Database Syst Rev, vol. 8, p. CD008570, 2012

    work page 2012

  11. [11]

    Proactive Selective Inhibition Targeted at the Neck Muscles: This Proximal Constraint Facilitates Learning and Regulates Global Control,

    I. D. Loram, B. Bate, P. Harding, R. Cunningham, and A. Loram, “Proactive Selective Inhibition Targeted at the Neck Muscles: This Proximal Constraint Facilitates Learning and Regulates Global Control,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 4320, no. 4, pp. 357–369, 2017

    work page 2017

  12. [12]

    Real-Time Ultrasound Segmentation, Analysis and Visualisation of Deep Cervical Muscle Structure,

    R. J. Cunningham, P. J. Harding, and I. D. Loram, “Real-Time Ultrasound Segmentation, Analysis and Visualisation of Deep Cervical Muscle Structure,” IEEE Trans. Med. Imaging, vol. 36, no. 2, pp. 653– 665, Feb. 2017

    work page 2017

  13. [13]

    Structure and function of the skeletal muscle extracellular matrix,

    A. R. Gillies and R. L. Lieber, “Structure and function of the skeletal muscle extracellular matrix,” Muscle Nerve, vol. 44, no. 3, pp. 318–31, 2011

    work page 2011

  14. [14]

    Correlation between active and passive isometric force and intramuscular pressure in the isolated rabbit tibialis anterior muscle,

    J. Davis, K. R. Kaufman, and R. L. Lieber, “Correlation between active and passive isometric force and intramuscular pressure in the isolated rabbit tibialis anterior muscle,” J. Biomech., vol. 36, no. 4, pp. 505–512, 2003

    work page 2003

  15. [15]

    Synchronous MRI of muscle motion induced by electrical stimulation,

    X. Deligianni et al., “Synchronous MRI of muscle motion induced by electrical stimulation,” Magn. Reson. Med., vol. 77, no. 2, pp. 664–672, 2017

    work page 2017

  16. [16]

    Electromyography: some methodological problems and issues,

    K. Türker, “Electromyography: some methodological problems and issues,” Phys. Ther., vol. 73, no. 10, pp. 698–710, 1993

    work page 1993

  17. [17]

    Neuromuscular ultrasound for evaluation of the diaphragm,

    A. Sarwal, F. O. Walker, and M. S. Cartwright, “Neuromuscular ultrasound for evaluation of the diaphragm,” Muscle and Nerve, vol. 47, no. 3. pp. 319–329, 2013

    work page 2013

  18. [18]

    Comb-push Ultrasound Shear Elastography (CUSE): A novel and fast technique for shear elasticity imaging,

    P. Song, M. W. Urban, A. Manduca, H. Zhao, J. F. Greenleaf, and S. Chen, “Comb-push Ultrasound Shear Elastography (CUSE): A novel and fast technique for shear elasticity imaging,” in IEEE International Ultrasonics Symposium, IUS, 2012, pp. 1842–1845

    work page 2012

  19. [19]

    Elastography for Muscle Biomechanics: Toward the Estimation of Individual Muscle Force,

    F. Hug, K. Tucker, J. L. Gennisson, M. Tanter, and A. Nordez, “Elastography for Muscle Biomechanics: Toward the Estimation of Individual Muscle Force,” Exerc. Sport Sci. Rev., vol. 43, no. 3, pp. 125–133, 2015

    work page 2015

  20. [20]

    Shear wave elastography reveals different degrees of passive and active stiffness of the neck extensor muscles,

    A. V. Dieterich et al., “Shear wave elastography reveals different degrees of passive and active stiffness of the neck extensor muscles,” Eur. J. Appl. Physiol., vol. 117, no. 1, pp. 171–178, 2016

    work page 2016

  21. [21]

    Current status of musculoskeletal application of shear wave elastography,

    J. Ryu and W. K. Jeong, “Current status of musculoskeletal application of shear wave elastography,” Ultrasonography, vol. 36, no. 3, pp. 185–197, 2017

    work page 2017

  22. [22]

    Validation of shear wave elastography in skeletal muscle,

    S. F. Eby, P. Song, S. Chen, Q. Chen, J. F. Greenleaf, and K. N. An, “Validation of shear wave elastography in skeletal muscle,” J. Biomech., vol. 46, no. 14, pp. 2381–2387, 2013

    work page 2013

  23. [23]

    Measurement of muscle contraction with ultrasound imaging,

    P. W. Hodges, L. H. M. Pengel, R. D. Herbert, and S. C. Gandevia, “Measurement of muscle contraction with ultrasound imaging,” Muscle Nerve, vol. 27, no. 6, p. 682, 2003

    work page 2003

  24. [24]

    Use of ultrasound to make noninvasive in vivo measurement of continuous changes in human muscle contractile length,

    I. D. Loram, C. N. Maganaris, and M. Lakie, “Use of ultrasound to make noninvasive in vivo measurement of continuous changes in human muscle contractile length,” J. Appl. Physiol., vol. 100, no. 4, pp. 1311– 1323, 2005

    work page 2005

  25. [25]

    Assessment of muscle fatigue using sonomyography: Muscle thickness change detected from ultrasound images,

    J. Shi, Y. P. Zheng, X. Chen, and Q. H. Huang, “Assessment of muscle fatigue using sonomyography: Muscle thickness change detected from ultrasound images,” Med. Eng. Phys., vol. 29, no. 4, pp. 472–479, 2007

    work page 2007

  26. [26]

    Predictability of in vivo changes in pennation angle of human tibialis anterior muscle from rest to maximum isometric dorsiflexion,

    C. N. Maganaris and V. Baltzopoulos, “Predictability of in vivo changes in pennation angle of human tibialis anterior muscle from rest to maximum isometric dorsiflexion,” Eur. J. Appl. Physiol. Occup. Physiol., vol. 79, no. 3, pp. 294–297, 1999

    work page 1999

  27. [27]

    Automated regional analysis of B-mode 11 ultrasound images of skeletal muscle movement,

    J. Darby, E. F. E. F. Hodson-Tole, N. Costen, and I. D. I. D. Loram, “Automated regional analysis of B-mode 11 ultrasound images of skeletal muscle movement,” J. Appl. Physiol., vol. 112, no. 2, pp. 313–327, 2011

    work page 2011

  28. [28]

    Estimation of longitudinal muscle motion using a primal-dual algorithm,

    J. Z. Li, Y. Zhou, Y. F. Yang, B. Chen, L. Wang, and Y. P. Zheng, “Estimation of longitudinal muscle motion using a primal-dual algorithm,” in 5th 2012 Biomedical Engineering International Conference, BMEiCON 2012, 2012

    work page 2012

  29. [29]

    Estimating skeletal muscle fascicle curvature from B-mode ultrasound image sequences,

    J. Darby, B. Li, N. Costen, I. Loram, and E. Hodson- Tole, “Estimating skeletal muscle fascicle curvature from B-mode ultrasound image sequences,” IEEE Trans. Biomed. Eng., vol. 60, no. 7, pp. 1935–1945, 2013

    work page 1935

  30. [30]

    Reliability and validity of ultrasound measurements of muscle fascicle length and pennation in humans: a systematic review,

    L. K. Kwah, R. Z. Pinto, J. Diong, and R. D. Herbert, “Reliability and validity of ultrasound measurements of muscle fascicle length and pennation in humans: a systematic review,” J. Appl. Physiol., vol. 114, no. January 2013, pp. 761–769, 2013

    work page 2013

  31. [31]

    3D fascicle orientations in triceps surae,

    M. Rana, G. Hamarneh, and J. M. Wakeling, “3D fascicle orientations in triceps surae,” J. Appl. Physiol., vol. 115, no. 1, pp. 116–125, 2013

    work page 2013

  32. [32]

    Changes in the length and three-dimensional orientation of muscle fascicles and aponeuroses with passive length changes in human gastrocnemius muscles,

    R. D. Herbert, M. E. Heroux, J. Diong, L. E. Bilston, S. C. Gandevia, and G. A. Lichtwark, “Changes in the length and three-dimensional orientation of muscle fascicles and aponeuroses with passive length changes in human gastrocnemius muscles,” J. Physiol., 2014

    work page 2014

  33. [33]

    Transverse Strains in Muscle Fascicles during Voluntary Contraction: A 2D Frequency Decomposition of B-Mode Ultrasound Images,

    J. M. Wakeling and A. Randhawa, “Transverse Strains in Muscle Fascicles during Voluntary Contraction: A 2D Frequency Decomposition of B-Mode Ultrasound Images,” Int. J. Biomed. Imaging, vol. 2014, p. 352910, 2014

    work page 2014

  34. [34]

    Deep residual networks for quantification of muscle fiber orientation and curvature from ultrasound images,

    R. J. Cunningham, P. Harding, and I. D. Loram, “Deep residual networks for quantification of muscle fiber orientation and curvature from ultrasound images,” in Communications in Computer and Information Science, vol. 723, Springer, Cham, 2017, pp. 63–73

    work page 2017

  35. [35]

    In-vivo determination of 3D muscle architecture of human muscle using free hand ultrasound,

    M. Rana and J. M. Wakeling, “In-vivo determination of 3D muscle architecture of human muscle using free hand ultrasound,” J. Biomech., vol. 44, no. 11, pp. 2129–2135, 2011

    work page 2011

  36. [36]

    Estimating Full Regional Skeletal Muscle Fibre Orientation from B-Mode Ultrasound Images Using Convolutional, Residual, and Deconvolutional Neural Networks,

    R. Cunningham, M. Sánchez, G. May, and I. Loram, “Estimating Full Regional Skeletal Muscle Fibre Orientation from B-Mode Ultrasound Images Using Convolutional, Residual, and Deconvolutional Neural Networks,” J. Imaging, vol. 4, no. 2, p. 29, Jan. 2018

    work page 2018

  37. [37]

    Muscle force per cross-sectional area is inversely related with pennation angle in strength trained athletes.,

    S. Ikegawa, K. Funato, N. Tsunoda, H. Kanehisa, T. Fukunaga, and Y. Kawakami, “Muscle force per cross-sectional area is inversely related with pennation angle in strength trained athletes.,” J. Strength Cond. Res., vol. 22, no. 1, pp. 128–31, 2008

    work page 2008

  38. [38]

    Using ultrasound images of the forearm to predict finger positions,

    C. Castellini, G. Passig, and E. Zarka, “Using ultrasound images of the forearm to predict finger positions,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 20, no. 6, pp. 788–797, 2012

    work page 2012

  39. [39]

    A realistic implementation of ultrasound imaging as a human- machine interface for upper-limb amputees,

    D. Sierra Gonzalez and C. Castellini, “A realistic implementation of ultrasound imaging as a human- machine interface for upper-limb amputees,” Front. Neurorobot., vol. 7, p. 17, 2013

    work page 2013

  40. [40]

    Measurement of fibre pennation using ultrasound in the human quadriceps in vivo,

    O. M. Rutherford and D. A. Jones, “Measurement of fibre pennation using ultrasound in the human quadriceps in vivo,” Eur. J. Appl. Physiol. Occup. Physiol., vol. 65, no. 5, pp. 433–437, 1992

    work page 1992

  41. [41]

    Muscle cross- sectional area and voluntary force production characteristics in elite strength- and endurance-trained athletes and sprinters,

    K. Häkkinen and K. L. Keskinen, “Muscle cross- sectional area and voluntary force production characteristics in elite strength- and endurance-trained athletes and sprinters,” Eur. J. Appl. Physiol. Occup. Physiol., vol. 59, no. 3, pp. 215–220, 1989

    work page 1989

  42. [42]

    Change in length of relaxed muscle fascicles and tendons with knee and ankle movement in humans,

    R. D. Herbert, A. M. Moseley, J. E. Butler, and S. C. Gandevia, “Change in length of relaxed muscle fascicles and tendons with knee and ankle movement in humans,” J. Physiol., vol. 539, no. 2, p. 637, 2002

    work page 2002

  43. [43]

    A new method for measuring passive length-tension properties of human gastrocnemius muscle in vivo,

    P. D. Hoang, R. B. Gorman, G. Todd, S. C. Gandevia, and R. D. Herbert, “A new method for measuring passive length-tension properties of human gastrocnemius muscle in vivo,” J. Biomech., vol. 38, no. 6, pp. 1333–1341, 2005

    work page 2005

  44. [44]

    Estimation of Muscle Fiber Orientation in Ultrasound Images Using Revoting Hough Transform (RVHT),

    Y. Zhou and Y. P. Zheng, “Estimation of Muscle Fiber Orientation in Ultrasound Images Using Revoting Hough Transform (RVHT),” Ultrasound Med. Biol., vol. 34, no. 9, pp. 1474–1481, Sep. 2008

    work page 2008

  45. [45]

    Automated tracking of muscle fascicle orientation in B-mode ultrasound images,

    M. Rana, G. Hamarneh, and J. M. Wakeling, “Automated tracking of muscle fascicle orientation in B-mode ultrasound images,” J. Biomech., vol. 42, no. 13, pp. 2068–2073, 2009

    work page 2068

  46. [46]

    Automatic tracking of medial gastrocnemius fascicle length during human locomotion,

    N. J. Cronin, C. P. Carty, R. S. Barrett, and G. A. Lichtwark, “Automatic tracking of medial gastrocnemius fascicle length during human locomotion,” J. Appl. Physiol., 2011

    work page 2011

  47. [47]

    Computational methods for quantifying in vivo muscle fascicle curvature from ultrasound images,

    A. I. L. Namburete, M. Rana, and J. M. Wakeling, “Computational methods for quantifying in vivo muscle fascicle curvature from ultrasound images,” J. Biomech., vol. 44, no. 14, pp. 2538–2543, 2011

    work page 2011

  48. [48]

    Multilevel and Motion Model-Based Ultrasonic Speckle Tracking Algorithms,

    F. Yeung, S. F. Levinson, and K. J. Parker, “Multilevel and Motion Model-Based Ultrasonic Speckle Tracking Algorithms,” Ultrasound Med. Biol., vol. 24, no. 3, pp. 427–441, Mar. 1998

    work page 1998

  49. [49]

    Feature-Adaptive Motion Tracking of Ultrasound Image Sequences Using A Deformable Mesh,

    F. Yeung, S. F. Levinson, D. Fu, and K. J. Parker, “Feature-Adaptive Motion Tracking of Ultrasound Image Sequences Using A Deformable Mesh,” vol. 17, no. 6, pp. 945–956, 1998

    work page 1998

  50. [50]

    Recognaizing Handwritten Digital Using Hierarchical Products of Experts,

    G. Mayraz and G. E. Hinton, “Recognaizing Handwritten Digital Using Hierarchical Products of Experts,” vol. 24, 2, no. 2, pp. 189–197, 2002

    work page 2002

  51. [51]

    Deep learning,

    Y. LeCun, Y. Bengio, G. Hinton, L. Y., B. Y., and H. G., “Deep learning,” Nature, vol. 521, no. 7553, pp. 436–444, 2015

    work page 2015

  52. [52]

    Auto-Encoding Variational Bayes,

    D. P. Kingma and M. Welling, “Auto-Encoding Variational Bayes,” no. Ml, pp. 1–14, 2013

    work page 2013

  53. [53]

    Handwritten Digit Recognition with a Back-Propagation Network,

    L. D. Le Cun Jackel, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, B. Le Cun, J. Denker, and D. Henderson, “Handwritten Digit Recognition with a Back-Propagation Network,” Adv. Neural Inf. Process. Syst., pp. 396–404, 1990

    work page 1990

  54. [54]

    ImageNet Classification with Deep Convolutional Neural Networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” Adv. Neural Inf. Process. Syst. 25, pp. 1–9, 2012

    work page 2012

  55. [55]

    Deep Residual Learning for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learning for Image Recognition,” Arxiv.Org, vol. 7, no. 3, pp. 171–180, 2015. 12

    work page 2015

  56. [56]

    Rectified Linear Units Improve Restricted Boltzmann Machines,

    V. Nair and G. E. Hinton, “Rectified Linear Units Improve Restricted Boltzmann Machines,” Proc. 27th Int. Conf. Mach. Learn., no. 3, pp. 807–814, 2010

    work page 2010

  57. [57]

    Dropout : A Simple Way to Prevent Neural Networks from Overfitting,

    G. Hinton, “Dropout : A Simple Way to Prevent Neural Networks from Overfitting,” vol. 15, pp. 1929– 1958, 2014

    work page 1929

  58. [58]

    Learning Deconvolution Network for Semantic Segmentation,

    H. Noh, S. Hong, and B. Han, “Learning Deconvolution Network for Semantic Segmentation,” in 2015 IEEE International Conference on Computer Vision (ICCV), 2015, vol. 11-18-Dece, pp. 1520–1528

    work page 2015

  59. [59]

    Training products of experts by minimizing contrastive divergence,

    G. E. Hinton, “Training products of experts by minimizing contrastive divergence,” Neural Comput., vol. 14, no. 8, pp. 1771–1800, 2002

    work page 2002

  60. [60]

    Early Visual Concept Learning with Unsupervised Deep Learning,

    I. Higgins et al., “Early Visual Concept Learning with Unsupervised Deep Learning,” 2016

    work page 2016

  61. [61]

    Information processing in dynamical systems: Foundations of harmony theory,

    P. Smolensky, “Information processing in dynamical systems: Foundations of harmony theory,” Parallel Distrib. Process. Explor. Microstruct. Cogn., vol. 1, no. 1, pp. 194–281, 1986

    work page 1986

  62. [62]

    The ‘wake-sleep’ algorithm for unsupervised neural networks.,

    G. E. Hinton, P. Dayan, B. J. Frey, and R. M. Neal, “The ‘wake-sleep’ algorithm for unsupervised neural networks.,” Science (80-. )., vol. 268, no. 5214, pp. 1158–1161, 1995

    work page 1995

  63. [63]

    Learning sets of filters using back-propagation,

    D. C. Plaut and G. E. Hinton, “Learning sets of filters using back-propagation,” Comput. Speech Lang., vol. 2, no. 1, pp. 35–61, 1987

    work page 1987

  64. [64]

    Stacked convolutional auto-encoders for hierarchical feature extraction,

    J. Masci, U. Meier, D. Cireşan, and J. Schmidhuber, “Stacked convolutional auto-encoders for hierarchical feature extraction,” Lect. Notes Comput. Sci. (including Subser. Lect. Notes Artif. Intell. Lect. Notes Bioinformatics), vol. 6791 LNCS, no. PART 1, pp. 52–59, 2011

    work page 2011

  65. [65]

    Reducing the Dimensionality of Data with Neural Networks\r,

    G. E. Hinton and R. R. Salakhutdinov, “Reducing the Dimensionality of Data with Neural Networks\r,” Science (80-. )., vol. 313, no. 5786, pp. 504–507, 2006

    work page 2006

  66. [66]

    Generative Adversarial Networks,

    I. Goodfellow, J. Pouget-Abadie, and M. Mirza, “Generative Adversarial Networks,” arXiv Prepr. arXiv …, pp. 1–9, 2014

    work page 2014

  67. [67]

    Imagenet classification with deep convolutional neural networks,

    a Krizhevsky, I. Sutskever, and G. Hinton, “Imagenet classification with deep convolutional neural networks,” Adv. Neural Inf. Process. Syst., pp. 1097– 1105, 2012

    work page 2012

  68. [68]

    Deep Neural Networks for Acoustic Modeling in Speech Recognition,

    G. Hinton et al., “Deep Neural Networks for Acoustic Modeling in Speech Recognition,” IEEE Signal Process. Mag., no. November, pp. 82–97, 2012

    work page 2012

  69. [69]

    Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification,

    K. He, X. Zhang, S. Ren, and J. Sun, “Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification,” CoRR, vol. abs/1502.0, 2015

    work page 2015

  70. [70]

    FlowNet: Learning Optical Flow with Convolutional Networks,

    P. Fischer, A. Dosovitskiy, E. Ilg, ..., and T. Brox, “FlowNet: Learning Optical Flow with Convolutional Networks,” Iccv, p. 8, 2015

    work page 2015

  71. [71]

    The application of deep convolutional neural networks to ultrasound for modelling of dynamic states within human skeletal muscle,

    R. J. Cunningham, P. J. Harding, and I. D. Loram, “The application of deep convolutional neural networks to ultrasound for modelling of dynamic states within human skeletal muscle,” bioRxiv preprint https://doi.org/10.1101/157479, 2017

    work page doi:10.1101/157479 2017

  72. [72]

    Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs),

    D.-A. Clevert, T. Unterthiner, and S. Hochreiter, “Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs),” pp. 1–14, 2015

    work page 2015

  73. [73]

    Recruitment of motor units in the medial gastrocnemius muscle during human quiet standing: is recruitment intermittent? What triggers recruitment?,

    T. M. M. T. M. M. Vieira, I. D. I. D. Loram, S. Muceli, R. Merletti, and D. Farina, “Recruitment of motor units in the medial gastrocnemius muscle during human quiet standing: is recruitment intermittent? What triggers recruitment?,” J. Neurophysiol., vol. 107, no. 2, pp. 666–676, 2012

    work page 2012

  74. [74]

    Postural activation of the human medial gastrocnemius muscle: are the muscle units spatially localised?,

    T. M. M. M. Vieira, I. D. D. Loram, S. Muceli, R. Merletti, and D. Farina, “Postural activation of the human medial gastrocnemius muscle: are the muscle units spatially localised?,” J. Physiol., vol. 589, no. Pt 2, pp. 431–443, 2011

    work page 2011

  75. [75]

    Active Shape Models-Their Training and Application,

    T. F. Cootes, C. J. Taylor, D. H. Cooper, and J. Graham, “Active Shape Models-Their Training and Application,” Computer Vision and Image Understanding, vol. 61, no. 1. pp. 38–59, 1995

    work page 1995

  76. [76]

    Signature Verification using a "Siamese" Time Delay Neural Network,

    J. Bromley, I. Guyon, Y. Lecun, E. Sickinger, and R. Shah, “Signature Verification using a "Siamese" Time Delay Neural Network,” in Advances in Neural Information Processing Systems, 1994

    work page 1994

  77. [77]

    Adam: A Method for Stochastic Optimization,

    D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimization,” pp. 1–15, 2014

    work page 2014

  78. [78]

    The application of b-mode ultrasonography for analysis of human skeletal muscle,

    R. Cunningham, “The application of b-mode ultrasonography for analysis of human skeletal muscle,” Manchester Metropolitan University, Manchester, England, UK, 2015

    work page 2015

  79. [79]

    A technical note on variable inter-frame interval as a cause of non-physiological experimental artefacts in ultrasound,

    D. Miguez, E. F. Hodson-Tole, I. Loram, and P. J. Harding, “A technical note on variable inter-frame interval as a cause of non-physiological experimental artefacts in ultrasound,” R. Soc. Open Sci., vol. 4, no. 5, 2017

    work page 2017

  80. [80]

    Bayesian GAN,

    Y. Saatchi and A. G. Wilson, “Bayesian GAN,” no. Nips, pp. 1–10, 2017

    work page 2017

Showing first 80 references.