Vous êtes ici : Accueil » Kiosque » Annonce


Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?


19 novembre 2020

Deep Learning for Muscle Characterisation

Catégorie : Stagiaire

With aging, several functional and anatomical impairments appear in the neuromuscular system and leads to the decline of the motor abilities. This motor decline can accelerate frailty appearance and potentially increase the fall risk. In addition to aging, inactivity and lack of physical activity have been identified as potential factors of the acceleration of sarcopenia. The positive influence of physical activity interventions on sarcopenia has been previously described [1]. Indeed, decreased physical activity that occurs with aging or bad daily-life habits may contribute to age-related sarcopenia.


Artificial intelligence (AI) has been developing rapidly in recent years in terms of algorithms, hardware implementation, and applications in a vast number of areas. Recently, AI has been widely used in applications such as disease diagnostics, living assistance, biomedical information processing, and biomedical research. It can be asserted that, just like AI itself, the application of AI in biomedical field is still in its early stage. New progress and breakthroughs will continue to push the frontier and widen the scope of AI application, and fast developments are envisioned in the near future.

The main goal of the proposed Master thesis is to introduce novel and efficient deep learning techniques for muscle aging evaluation by analyzing the recorded data from the muscles thanks to the huge progress in sensors technology. The advantage of data analysis is that the decisions can be based on knowledge gained from facts and thus to become less dependent on intuition and subjective experiences. With this knowledge, we will have a better interpretation and prediction of the aging. In addition to the latter, we can also get new insights about the effect of the different daily-life habits on the aging process. In order to evaluate the aging, the recorded data obtained from a sensor network grid namely high-density surface electromyogram (HD-sEMG) is analysed. It represents a noninvasive technique for measuring electrical neuromuscular activity. This technique has been extensively used in a variety of applications such as the control of prosthetic devices for individuals with amputations or congenitally deficient limbs, for the estimation of nerve conduction variable and anatomical properties of the muscular tissue [2].

Deep learning represents a growing and intensive research area with the goal of end-to-end system. It proceeds by giving raw data as input and by stacking more than the usual two neural layers. Each low- level layer encodes specific properties of the input data as primitives that are gradually combined by successive high-level layers in order to produce representative and hopefully discriminative features [4]. Currently, deep models are one of the leading and state of the art models in machine learning. They have been applied in many different domains. Most successful deep neural models are the ones with many layers which highly increases their number of parameters. Training such models requires a large number of training samples which we either hold from simulated data from Carriou et al. [3] model and recorded data from the European project (EIT Health CHRONOS 2018, lead: SU, co-lead: UTC) which has been granted to develop an innovative tool for the monitoring of muscle aging. In order to further boost the performances, prevent the network from overfitting and improve its generalization performance, we further intend to use data augmentation, early stopping, parameters sharing, unsupervised learning, dropout, batch normalization, etc. [4]

[1] EVANS, William J. Skeletal muscle loss: cachexia, sarcopenia, and inactivity. The American journal of clinical nutrition, 2010, vol. 91, no 4, p. 1123S-1127S.

[2] Al Harrach, M., Boudaoud, S., Carriou, V., Laforet, J., Letocart, A. J., Grosset, J. F., & Marin, F. (2017). Investigation of the HD-sEMG probability density function shapes with varying muscle force using data fusion and shape descriptors. Computers in biology and medicine, 89, 44-58

[3] CARRIOU, Vincent. Multiscale, multiphysic modeling of the skeletal muscle during isometric contraction. 2017. Thèse de doctorat. Compiègne.

[4] RIDA, Imad, AL-MAADEED, Noor, AL-MAADEED, Somaya, et al. A comprehensive overview of feature representation for biometric recognition. Multimedia Tools and Applications, 2018, p. 1-24.


- The candidate must be at the second year Master degree or 5th year of an engineering school in France (computer science, electrical engineering, machine learning, applied mathematics, data science)

- Background in machine learning, signal processing, deep learning and optimization

- Good coding skills for numerical simulation (Python, Pytorch, TensorFlow, Matlab, ...)

- Demonstrated analytical, verbal, and scientific writing skills in English


Please send your CV, cover letter, transcripts and recommendation letters to:

Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2020.