MU: PhD position – Three-dimensional texture analysis of ultrasound images applied to muscles and tendons

Download the thesis proposal

Contact

  • Thesis supervisor: Christine CHAPPARD – christine.chappard[at]inserm.fr

Duration

3 years – November 2026 – October 2029

Context

The use of ultrasound imaging in the musculoskeletal and articular fields has developed considerably over the last 20 years. There are 4 possible approaches: ultrasound B-mode imaging, doppler, elastography, and quantitative imaging based on physics such as velocity, attenuation and backscatter (Gutiérrez-Martinez J, 2019). Ultrasound B-mode imaging provides access to anatomical structures with a semiological description of the observed various lesions, and has benefited in recent years from improved image quality in terms of contrast and resolution (Schousboe JT, 2013).

Ultrasound B-mode is an interface imaging technique, each corresponding to a variation in impedance. When ultrasound interacts with small structures smaller than the wavelength, scattering appears on the ultrasound B image, called speckle, giving the image a textured appearance. The intensity of speckle echoes or Echo-Intensity (EI) are represented by gray levels that can be statistically analyzed. Statistical methods of order 1 only concern one pixel. It is then possible to calculate mean, median, standard deviation, skewness or kurtosis and entropy. Texture analysis of order 2 or higher describes the spatial organization of pixels in an image. Local image properties, periodicity phenomena and the identification of particular patterns can all be highlighted by texture analysis. These methods are widely used in satellite exploration to define the existence of reliefs, forests, fields, lakes, etc. (Hall beyer M, 2017). Its interest also lies in its ability to analyze low-resolution images (Hall beyer M, 2017). The advantage of second-order texture parameters is that they are invariant not depending on grayscale intensity, but only on grayscale variation into the image. The most popular methods are neighborhood-based ones, quantifying the differences in image grayscale intensity between 2 (order 2) or more pixels (order n). Run length (RL) and co-occurrence matrices (GLCM) belong to this family. Other approaches of textural analysis are less frequently used such as Local Binary Pattern (LBP), Fractals, Blobs, wavelets.

In medicine, texture analysis is highly developed in the field of oncology to differentiate benign and malignant tumors on images of different nature (MRI, PET, ultrasound) (Varghese BA, 2023).

Muscle function depends on muscle length, thickness and intrinsic quality, which in turn depends on the amount of intramuscular fat, the presence of fibrotic connective tissue or infiltration (Adkins A, 2020). Most ultrasound studies use visual scales defining particular patterns by identifying the more or less hypoechoic and homogeneous nature of the muscle structure (Helmy H, 2018). Quantitative ultrasound are always performed in two dimensions (2D) on a rectangular region of interest not taking account the muscle anatomy. In a recent review, muscle thickness measurement (37 papers) and muscle cross-section measurements (17 papers) were tested (Casey P, 2022). Fiber length and pennation angle are also explored, particularly in sarcopenia (Strasser EM, 2013). Muscle exploration based on EI and texture analysis of order ≥2 (38 papers) are more recently tested (Jan YK, 2025). Muscle thickness measurement is the only metric validated for sarcopenia diagnosis (Cruz-Jentoft AJ, 2019).

The main interest of textural analysis is the possible relationship with muscle quality. Indeed, it has been found to be related to effect of age (Watanabe T, 2017), muscular function (Rios Diaz J, 2019), muscle strength (Bell KE, 2022, Wilkinson TJ, 2021). Moreover, the radiomic strategy seems particularly relevant, indeed, the combination of 4 parameters resulted in 100% correct classification to differentiate muscle pathologies (myopathic, neurogenic) from controls (Sogawa K, 2017).

In view of exploration of tendons using quantitative ultrasound methods is still highly confidential. The most widely used methods are elastography, which, at a consensus conference, was not recommended as a substitute for B ultrasound for tendon exploration due to insufficient sensitivity and reproducibility (Sconfienza LM, 2018).

Considering the dimensional (3D) quantitative ultrasound analysis, only EI was evaluated in muscular dystrophies (de Jong HJ, 2023). Indeed, the 3D approach provides a better account of the complexity of internal structure and contours, but requires greater computing and storage capacities. Inter-operator variability and reproducbility of the whole chain of measurements remain as major problems in the parametric use of ultrasound, which can be partially overcome by 3D imaging. At this time, it is possible to obtain a 3D stable image during the scan, using optical, electro-magnetic, mechanical or even acoustic tracking systems.

Objective

Our goal is to develop a three-dimensional (3D) quantitative analysis on B-mode ultrasound muscle’s images and to compare our biomarkers with performance in three settings: among elite athletes, older adults with sarcopenia, and patients with chronic low back pain.

  • Study 1: Sport and performance
    The gastronecmius and quadriceps muscles are particularly sollicitated in competitive cyclists. This population exhibits remarkable strength capacity at high speeds and is therefore a good example for analyzing structural variations in the muscles of the lower limbs. Maximum strength and power levels tests will help to determine the maximum capabilities of the lower limbs and will be compared to 3D echo-B quantitative analyses of muscles.
  • Study 2: Sarcopenia in a geriatric population
    In a geriatric population, we have 3 objectives. The first objective is to test the reproducibility of 3D quantitative analysis on B-mode ultrasound images specific to this age group (over 75 years) at the rectus femoris and biceps brachii muscles. Our second objective is to evaluate the diagnostic performance of quantitative parameters derived from muscle ultrasound, potentially in combination, for diagnosing sarcopenia in older adults versus the gold standard which is the bioelectrical impedance analysis measurement according to EWGSOP (Cruz-Jentoft AJ, 2019). The third objective is to evaluate the predictive value of muscle ultrasound parameters versus ultrasound-based sarcopenia status according to EWGSOP 2 criteria for predicting adverse events (decline in physical performance, falls, fractures, dependency, mortality).
  • Study 3: Chronic lombalgia
    Low back pain is a major public health issue with a significant societal impact due to physical disability and socio-professional repercussions. Several studies have demonstrated a negative association between the cross sectional area of the spinal extensor muscles assessed by ultrasound and spinal muscle performance indicators such as the Sorensen test (Heidari P, 2015). We aim to compare 3D quantitative measurements from 3D echo-B ultrasound with clinical strength back muscle measurements (Sorensen test) and MRI.

Bibliography

  • Gutiérrez-Martínez J, et al. Clin Rheumatol. 2019. doi: 10.1007/s10067-019-04791-z. Review
  • Schousboe JT, et al. J Clin Densitom. 2013 doi: 10.1016/j.jocd.2013.08.004.
  • M Hall-Beyer. International Journal of Remote Sensing, 2017. https://doi.org/10.1080/01431161.2016.1278314
  • Varghese BA, et al. Front Radiol. 2023. doi: 10.3389/fradi.2023.124054
  • Adkins AN, Murray WM. J Vis Exp. 2020 doi: 10.3791/61765.
  • Helmy H, et al. Egypt J Neurol Psychiatr Neurosurg. 2018 doi: 10.1186/s41983-018-0039-6.
  • Casey P, et al. J Cachexia Sarcopenia Muscle. 2022 doi: 10.1002/jcsm.13041.
  • Strasser EM, et al. Age (Dordr). 2013 doi: 10.1007/s11357-013-9517-z.
  • Jan YK, et al. Diagnostics 2025. doi: 10.3390/diagnostics15050524.
  • Watanabe T, J Ultrasound Med 2017. Doi : 10.7863/ultra.16.07054.
  • Ríos-Díaz J, et al. Eur Radiol. 2019 doi: 10.1007/s00330-018-5943-8.
  • Bell KE, et al. J Cachexia Sarcopenia Muscle. 2022 doi: 10.1002/jcsm.12957.
  • Wilkinson TJ, et al. Ultrason Imaging. 2021 doi: 10.1177/01617346211009788.
  • Sogawa K, et al. Radiology. 2017 doi: 10.1148/radiol.2016160826.
  • Sconfienza LM, et al. Eur Radiol. 2020 doi: 10.1007/s00330-019-06545-6.
  • De Jong L et al. Muscle and nerve, 2023. doi: 10.1002/mus.27943.
  • J Cruz-Jentoft A et al. Age Ageing 2019. doi: 10.1093/ageing/afy169
  • Heidari P, et al. Asian J Sports Med. 2015. doi: 10.5812/asjsm.23803. Epub 2015 Jan 19.

Mission(s)

The candidate will be required to analyze images, perform segmentation, to automate segmentation using AI methods. The candidate will also use and develop a software developed on Python at the LIB able to calculate the main textural parameters (GLCM, RL, LBP, fractals), the volume, the mean thickness, and implement new quantification analysis such as the pennation angle and the anisotropy on the whole muscle taking account its anatomy. The candidate will privilege a radiomic approach to find the best combinations of parameters. The same methodology applied on muscle could also be applied in tendons.

Skills

Skills in image analysis, Python, statistics, models, radiomics

Remuneration

Doctoral contract (ED 393)