LKLyes Kadem

Lyes Kadem

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Lyes Kadem has a total of 106 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • An In-Vitro simulator for investigating flow disturbances in the left ventricle during controlled external impact

    by Dorian Sweidy, Andrew Mathyssen, Kowsar Teimouri, Wei Wei, Wael Saleh, Lyes Kadem, Morgane Evin


  • Open Access

    A high-fidelity simulator for evaluation of hemodynamic response during cardiopulmonary resuscitation in hypogravity environments

    by Zoé Lord, Christian Andrade, Lawrence Leroux, Lyes Kadem

    With the emergence of long-duration space travel, space exploration missions pose a major concern due to the heightened risk of medical emergencies, such as sudden cardiac arrest. While several cardiopulmonary resuscitation (CPR) methods have been proposed for human spaceflight, their reliability and effectiveness remain uncertain, as these methods lack systematic evaluation through physiological metrics. To address this gap, a high-fidelity CPR simulator was developed to simulate blood circulation and deliver real-time hemodynamic feedback. Herein, we show that in normogravity, the CPR simulator generates compression-decompression waveforms that align with published animal and test bench studies. As an exploratory comparison, we also report relative differences in hemodynamic pressure observed between normogravity and hypogravity conditions. The findings highlight that internal physiological responses are critical for evaluating CPR effectiveness in hypogravity, with the CPR simulator serving as a plausible tool. The current study represents an initial step toward the validation of a gold standard CPR protocol and may contribute to the complex health challenges surrounding long-duration spaceflight.

  • EchoGym: A Digital Twin Framework for Robotic Echocardiography Using Two-Stage Generative AI

    by Ehsan Zakeri, Amanda Spilkin, Hanae Elmekki, Antonela Zanuttini, Wen-Fang Xie, Lyes Kadem, Jamal Bentahar, Philippe Pibarot

  • Deep Reinforcement Learning-Based Ultrasound Visual Servoing Scheme for Autonomous Robotic Echocardiography

    by Ehsan Zakeri, Amanda Spilkin, Hanae Elmekki, Antonela Zanuttini, Wen-Fang Xie, Lyes Kadem, Jamal Bentahar, Philippe Pibarot

    This study develops a deep reinforcement learning (DRL)-based ultrasound visual servoing (DBUVS) scheme for an autonomous robotic echocardiography system. The proposed approach aims to replace remote cardiac sonographers with a newly developed AI agent, Cardiac Sonographer Net (CS-Net), trained using DRL based on the Soft Actor-Critic (SAC) algorithm. To train CS-Net, a robotic echocardiography environment is simulated in GAZEBO simulator using a two-stage generative AI (GenAI) approach to produce highfidelity synthetic ultrasound images. In the first stage, multiple convolutional neural networks (CNNs) generate initial ultrasound images based on different parameter settings from a given probe pose. A fuzzy inference system (FIS) then fuses these images into a single low-quality representation. In the second stage, a super-resolution generative adversarial network with gradient penalty (SRGAN-GP) enhances image quality. Compared with low-quality images, the GenAI-based outputs show an 11.37 perceptual image patch similarity (LPIPS) and higher resolution (256W256 to 500W500 pixels), closely matching real ultrasound images. CS-Net is initially trained in the simulation environment and deployed on the real experimental robotic system with an ultrasound probe mounted on the end effector and a cardiac phantom for testing, using sim-to-real transfer learning. Experimental results demonstrate that the robotic echocardiography system powered by CS-Net performs autonomous scanning with higher accuracy and efficiency than the echocardiography remotely operated by the sonographer. Specifically, the system achieves faster convergence by reaching an image feature error norm of 0.176 in 25 seconds, compared with 0.253 in 50 seconds for remote operation.


  • Open Access

    Comprehensive review of reinforcement learning for medical ultrasound imaging

    by Hanae Elmekki, Saidul Islam, Ahmed Alagha, Hani Sami, Amanda Spilkin, Ehsan Zakeri, Antonela Mariel Zanuttini, Jamal Bentahar, Lyes Kadem, Wen-Fang Xie, et al.

    Medical Ultrasound (US) imaging has seen increasing demands over the past years, becoming one of the most preferred imaging modalities in clinical practice due to its affordability, portability, and real-time capabilities. However, it faces several challenges that limit its applicability, such as operator dependency, variability in interpretation, and limited resolution, which are amplified by the low availability of trained experts. This calls for the need of autonomous systems that are capable of reducing the dependency on humans for increased efficiency and throughput. Reinforcement Learning (RL) comes as a rapidly advancing field under Artificial Intelligence (AI) that allows the development of autonomous and intelligent agents through rewarded interactions with their environments. Several existing surveys on advancements in US imaging predominantly focus on partially autonomous AI solutions. However, none of these surveys explore the intersection between the stages of the US process and the recent advancements in RL solutions. To bridge this gap, this survey proposes a comprehensive taxonomy that integrates the stages of the US process with the RL development pipeline -including data preparation, problem formulation, simulation environment, RL training, validation and finetuning- and reviews current research efforts under this taxonomy. This work aims to highlight the potential of RL in building autonomous US solutions while identifying limitations and opportunities for further advancements in this field.


  • Open Access

    Comprehensive Review of Reinforcement Learning for Medical Ultrasound Imaging

    by Hanae Elmekki, Saidul Islam, Ahmed Alagha, Hani Sami, Amanda Spilkin, Ehsan Zakeri, Antonela Mariel Zanuttini, Jamal Bentahar, Lyes Kadem, Wen-Fang Xie, et al.

    Medical Ultrasound (US) imaging has seen increasing demands over the past years, becoming one of the most preferred imaging modalities in clinical practice due to its affordability, portability, and real-time capabilities. However, it faces several challenges that limit its applicability, such as operator dependency, variability in interpretation, and limited resolution, which are amplified by the low availability of trained experts. This calls for the need of autonomous systems that are capable of reducing the dependency on humans for increased efficiency and throughput. Reinforcement Learning (RL) comes as a rapidly advancing field under Artificial Intelligence (AI) that allows the development of autonomous and intelligent agents through rewarded interactions with their environments. Several existing surveys on advancements in US imaging predominantly focus on partially autonomous AI solutions. However, none of these surveys explore the intersection between the stages of the US process and the recent advancements in RL solutions. To bridge this gap, this survey proposes a comprehensive taxonomy that integrates the stages of the US process with the RL development pipeline -including data preparation, problem formulation, simulation environment, RL training, validation and finetuning- and reviews current research efforts under this taxonomy. This work aims to highlight the potential of RL in building autonomous US solutions while identifying limitations and opportunities for further advancements in this field.


  • Open Access

    CACTUS: An open dataset and framework for automated Cardiac Assessment and Classification of Ultrasound images using deep transfer learning

    by Hanae Elmekki, Ahmed Alagha, Hani Sami, Amanda Spilkin, Antonela Mariel Zanuttini, Ehsan Zakeri, Jamal Bentahar, Lyes Kadem, Wen-Fang Xie, Philippe Pibarot, et al.


  • Robust Deep Feature Ultrasound Image-Based Visual Servoing: Focus on Cardiac Examination

    by Ehsan Zakeri, Amanda Spilkin, Hanae Elmekki, Antonela Zanuttini, Lyes Kadem, Jamal Bentahar, Wen-Fang Xie, Philippe Pibarot

    This article introduces a robust deep feature ultrasound image-based visual servoing (UIBVS) technique for an ultrasound robot focusing on automatic cardiac examination. To this end, a convolutional neural network named ultrasound-cardiac-feature-net (UCF-Net) is developed, which is trained in a supervised manner to process ultrasound images and generate a set of six image features referred to as deep ultrasound image features. To enhance the robustness of UCF-Net against the variables that affect the ultrasound image quality, such as interaction normal force, scan depth, dynamic range, power, and gain, several datasets with different sets of parameters are gathered for training. Deep ultrasound image features enable an eye-in-hand robot to interact with the human body through UIBVS. To implement UIBVS, a filtered integral quasi-super-twisting algorithm (FIQSTA) is synthesized as the primary controller. Interaction force control is also considered within a hybrid vision/force control framework, providing compliance with the body and increasing the safety of the interaction. The proof of the robustness and stability of FIQSTA is also investigated. Experimental results on a cardiac phantom for four main views, i.e., parasternal short axis, parasternal long axis, subcostal, and apical four chambers views, and a trajectory passing through the main views demonstrate the feasibility of the proposed method for cardiac examination and the superior performance of the main controller to other well-known methods, including proportional (P) controller, sliding mode controller, super-twisting algorithm (STA), and integral quasi-STA.


  • Open Access

    Role of Non-Invasive Hemodynamic Forces through Four-Dimensional-Flow Magnetic Resonance Imaging (4D-Flow MRI) in Evaluating Mitral Regurgitation with Preserved Ejection Fraction: Seeking Novel Biomarkers

    by Monisha Ghosh Srabanti, Corey Adams, Lyes Kadem, Julio Garcia

    Mitral regurgitation (MR) is the systolic retrograde flow from the left ventricle (LV) to the left atrium. Despite the recognized importance of hemodynamic force (HDF) in cardiology, its exploration in MR has been limited. Therefore, we aimed to explore non-invasively assessed HDF as a novel biomarker for evaluating MR utilizing 4D-flow MRI. The study cohort comprised 15 healthy controls (19–61 years, 53% men) and 26 MR patients with preserved ejection fraction (EF) (33–75 years, trivial–severe, 54% men). The HDF analysis involved the semi-automatic calculation of systolic–diastolic root mean square (RMS), average, and transverse/longitudinal ratio across three directions (S-L: septal–lateral, I-A: inferior–anterior, and B-A: basal–apical) using Segment, v2.2 R6410 (Lund, Sweden, Medviso). A noticeable trend shift emerged in HDF as the MR severity increased (p-value < 0.05). The MR severity demonstrated a noteworthy correlation with systolic RMS B-A, average B-A, diastolic average B-A, systolic average S-L, B-A, and systolic–diastolic ratio (rho = 0.621, 0.457, 0.317, 0.318, 0.555, −0.543, −0.35, respectively; p-value < 0.05). HDF significantly correlated with LV function (end-diastolic volume, end-systolic volume, EF, and mass; p-value < 0.05). Systolic RMS B-A and diastolic RMS S-L emerged as significant predictors of MR (Beta, 95% CI [3.253, 1.204–5.301], [5.413, 0.227–10.6], p-value < 0.05). This study emphasizes HDF as a potential hemodynamic biomarker for evaluating MR.