ENDOSIM is a research project in the field of predictive simulation and computer-aided medical interventions (CAMI). It focuses on the treatment of aortic aneurisms and valvular stenoses. In a previous ANR TecSan project (ANGIOVISION, ended in February 2013), the partners of the ENDOSIM project have developed operative assistance tools using augmented angio-navigation for the treatment of abdominal aortic aneurysms (AAA). The results demonstrated, on more than 20 patients, the accuracy of the patient-specific simulation approach. Based on these developments and results, the team aims to move forward and tackle the problem of predictive planning, in order to maximize the accuracy and reliability of two complex endovascular procedures: • the implantation of fenestrated stent-grafts for the treatment of thoraco-abdominal aneurysms, • the endovascular implantation of cardiac valves for the treatment of aortic stenoses. For these two minimally invasive procedures, atheromatous plaques are sources of numerous, unsolved so far, difficulties among which: navigability issues in the vicinity of the lesions, risk of plaque rupture due to ancillary contacts, complexity for positioning the device on the lesion site, brittleness of the vasculature, crushing of the native valves… These issues currently constitute a major obstacle for a more massive use of endovascular techniques. The goal of ENDOSIM is to develop the first predictive endovascular surgery planning software in the world. This will lead to optimize the pre-operative planning and to secure per-operative navigation, through the following points: • tool navigability estimation from the patient’s imaging data, • improvement of the pre-operative device sizing reliability, • pre-operative prediction of the device positioning and per-operative visualization, • decision-making help for patient eligibility and device selection. In order to reach these objectives, the novel approach featured in ENDOSIM relies upon the joint use of image analysis techniques and biomechanical numerical simulation techniques, both being patient-specific and predictive. The scientific breakthroughs of ENDOSIM comprise mainly accurate and predictive patient-specific simulations of the endovascular ancillary insertion and device deployment. These simulations will be based on pre-operative imaging data and validated using per- and post-operative data on a group of atheromatous patients. The prediction of the risk of surgery-induced injury at the atheromatous sites is also very original. The numerical simulations developed through the project will be systematically enhanced and validated thanks to 3D imaging data obtained on real patients with the per-operative multi-incidence equipment of the TherA-Image platform. From a clinical point of view, the benefits of the ENDOSIM project will relate to securing the surgical planning thanks to simulations based on pre-operative data and improved positioning accuracy thanks augmented navigation tools. This should allow a more massive use of endovascular treatments and hence make the most of these minimally invasive procedures for the patients. From the industrial point of view, ENDOSIM will lead Therenva® (French leader in endovascular surgery software) to market the first predictive endovascular planning software solution. This will also be complemented by a visualization system for per-operative assistance. The close partnership with Ansys® (worldwide leader in numerical simulation) will promote a widespread adoption of Therenva® software solutions by endovascular device companies, as a first step, and by the worldwide clinical community as a second step.

The ACCORDS project aims at solving three issues related to Internet of Things (IoT). The first one consists in developing a software library to ensure interoperability with a set of sensors while putting aside the sensor model to focus on data. The current lack of interoperability restricts the number of sensors that are usable as part of an integrated-sensor platform. Sensors that do not meet the standard protocol ISO/IEEE 11073 (specific to quantified-self devices) are generally excluded from such platforms, despite the small market share of sensors based on standard protocols. Indeed, manufacturers would rather use proprietary protocols like GAFA (Google Apple Facebook Amazon) ones. Thus it underlines the need to communicate with all sensors, no matter which standard or proprietary protocol they use, in order to cover the whole on-market sensors categories. The second challenge is to measure sensors accuracy since most of them are not certified as medical devices. An experiment will be conducted on the best-selling sensors in order to assess their reliability and to compare their measures with standard ones. Studies on health sensors are usually based on the use of medical devices, so as to avoid dealing with potentially inaccurate data. The work carried out for the mainstream market is showing that, considering the example of connected wrists, the measure reliability is decreasing during moderated to high intensity exercises compared to the test at rest, or in case of varied and/or amplified body motion (walking, running, biking), and results are also user dependent (depending on the skin thickness, or motion pattern). Now, even if for some connected wrists, the measure of the number of steps seems quite reliable, the deducted crossed distance may be erroneous. As far as it is possible, to estimate the sensors' reliability experiments will be conducted to evaluate and compare the reliability of the most common sensors to gold-standards. This approach is responding to the increasing demand of consumers to get some guarantee on the results quality regarding the continuous development of more or less reliable wearable technologies. Thirdly, we will analyze health data interactions in a multimodal context for a better health follow-up, especially for the elderly. The final idea is to build, thanks to these three axes, a cloud where all data would be available, in association with their context, to the new health systems builders, namely states, regions, departments, cities, public or freelance establishments, software publisher, mutual insurances, allowing them to work at infinitesimal costs and avoiding current systems limitations. This cloud would be managed by a trusted third-party whose neutrality will enable each individual to register his own personal health data and to decide to share them or not, in whole or in part, temporarily or definitely, the user being at the heart of the decision-making process. This new regulatory paradigm with finally clarify the different sector operators roles: users, health professionals, industrials and insurance companies. Insuring readability, reliability, and security of multimodal data gathered in the user daily living space allows to support the development of health assisted self-management, which raises clinical, economic, and ethical issues, related to public health actions.

Robot-assisted surgery, allowing surgeons to perform complex surgeries through tiny incisions, has been significantly increasing in popularity worldwide. However, surgical safety is still a major concern in the high-risk operating environment and remains a threat to the quality of surgical outcome. As global statistics, millions of surgeries per year would encounter safety-critical intraoperative adverse events, most of which were otherwise avoidable if the surgeon can be timely aware of the potential risks in operation. In this project, we aim to introduce smart context-awareness into robot- assisted surgery, by developing novel artificial intelligence techniques to provide automatic cognitive assistance for surgeons during critical moments of the procedure, in order to improve surgical safety and quality. The use case of this project will be robot-assisted hysterectomy, which is the most common gynecological procedure performed on women diagnosed with uterine fibroids or cervical cancer. Both Hong Kong and French teams will explore together innovative multimodal machine learning methods, based on available synchronized clinical video and kinematic data, which will be more advanced and clinically relevant than all existing methods that only used visual perception. Based on our pilot studies, we have identified a set of critical intraoperative scenarios to address avoidable adverse events in hysterectomy, such as injury of the pelvic ureter during both the coagulation of the uterus pedicle and adnexectomy. To achieve our goal, we will solve the following key challenges: 1) How to yield precise and real-time recognition of the surgical context, i.e., surgical workflow, operation actions, surgical instruments, anatomical tissues and the reconstructed 3D surgical environments. 2) How to conduct automatic assessment of the identified critical-context-of-safety (CCS), and further provide informed decision-making support to surgeons for their best practice to avoid safety risks. By a research collaboration between world-class teams with complementary expertise and already-available clinical and annotated data, the i-SaferS project will generate outputs that provide fundamentally new and generic solutions and impactful references to the field. The project outcomes will significantly contribute to the emerging field of intelligent robotic surgery, and further strengthen the leading competitiveness of both partners in this field nationally and internationally.

Vagus nerve stimulation (VNS) is an approved clinical therapy for medically refractory epilepsy and depression. More recently, VNS has been proposed as a promising therapeutic approach for other pathologies such as heart failure, cardiac arrhythmia, inflammation and auto-immune diseases. One common difficulty currently encountered in all these established or promising clinical applications is to deliver an efficient therapy, while minimizing side effects. This is a particularly complex problem in the case of VNS, since a typical stimulation pattern consists of a set of biphasic pulses, characterized by several parameters, delivered through different electrode configurations. Moreover, the effects of VNS are poorly known and complex to study, since they involve many different organs and physiological functions and may change through time, due to neural or organ remodeling. Due to this complexity, current VNS technology is applied using fixed parameters obtained from limited, non-optimal manual titration and this simplistic approach may explain the lack of effect or the intolerance to the therapy. Although many efforts have been performed recently to propose closed-loop VNS methods, the proposed algorithms remain simple, limited to the modulation of one parameter and without a clear definition of the appropriate control variables. Furthermore, current electrode technologies provide a poor spatial resolution for stimulation and a low signal-to-noise ratio for neural recording, also limiting the development of advanced closed-loop approaches. We hypothesize that the use of an automated, closed-loop and subject-specific method for VNS parameter optimization, integrating new electrode technologies and new knowledge of the underlying physiology, may lead to an improved outcome for VNS patients and to address new therapeutic applications. The main objective of this project is thus to propose such novel data processing methods and electrode technologies, allowing for a closed-loop, subject-specific optimization of VNS therapy. Although the proposed methods and technologies will be generic, a second objective is to explore, through extensive in-silico, in-situ and in-vivo experimentations, the usefulness of the proposed system on a promising therapeutic target for VNS: the prevention of Sudden and Unexpected Death in Epileptic Patients (SUDEP). This second objective requires 1) the early detection of the potential occurrence of a SUDEP event and 2) the application of an original acute, adaptive VNS, to block the propagation through vagal efferent pathways, in order to prevent bradycardia and respiratory arrest. This project is organized in 6 work-packages. WP1 concerns project management. Two WP will be focused on technical developments: WP2 for data processing, modelling and control methods and WP3, focused on novel electrode technologies. A prototype neurostimulation system integrating these technologies will be developed in WP4. Finally, WP5 and WP6 will address the in-situ and in-vivo animal experimentations required for this project. AdaptVNS is an ambitious project, that we consider however feasible, on the basis of progress achieved in recent years by the consortium and on their available intellectual property (5 patents). Although this project presents some risks, it has a high potential of societal and industrial impact. If this project is successful, the final product will be a complete, functional, neuromodulation system prototype, integrating advanced closed-loop methods and organic electrode technologies. To our knowledge, there is no equivalent system today. Also, new ways for preventing SUDEP will be investigated. Such results may open new ways to optimally deliver VNS on current target clinical applications, to provide novel therapeutic functions, but also, to renew research on novel VNS therapies which are not always effective when delivered with standard technologies (heart failure, antiarrhythmic therapies, etc).

Cardiac resynchronization therapy (CRT) is an implant-based therapy applied to patients with a specific heart failure (HF) profile. The identification of CRT candidates is a challenging task. The application of current guidelines still induce a non-responder rate of about 30% and death remains high even after CRT implantation. A best selection of patients before implantation is essential to improve the individual quality of care and prevent the risk of non-justified complications. Recently, the assessment of left ventricular mechanics by speckle tracking echocardiography has been shown to provide useful information for patient selection and follow-up. Within the EXPERT project, explainable artificial intelligence (AI) methods, integrating machine-learning (ML) models and physiological in-silico models (patient digital twin), will be proposed to combine physiological knowledge with observed data, using model-based reasoning, to improve the interpretability of the approach while minimizing overfitting and limited robustness. Concerning methodological developments, novel patient-specific in-silico models of the cardiovsacular system will be proposed and will provide explainable model-based features that have a direct physiological meaning. Data-driven feature extraction will be also performed from clinical, electrocardiographic, and echocardiographic data. A hybrid modeling approach, which combines in-silico and ML models, will be proposed for the prediction of each patient response to a CRT intervention. These novel models should be evaluated clinically for the prediction of each patient response to a CRT intervention and to support the medical decision process for implanting or not a patient. The proposed hybrid classifier will be embedded in a novel decision support system (DSS) and will be used in inference mode to propose a new multivariate score, associated with an estimation of the probability of response. This approach will require the development of a technical architecture integrating all the available patient data and the calculation of a patient-specific probability of response in a timely manner Another objective of the EXPERT project will be to manage an original and unique pilot platform, for sharing a common secure and well labeled database of echocardiographic data. This scientific ambition bypasses the issues related to manufacturers and federates public partners, including three university hospitals with local servers localized in Rennes, in an open but sovereign initiative. This project could potentially serve the health data hub (HDH) initiative (specifically for echocardiography). The governance will be shared by the consortium members. All parts of the infrastructure and the whole project are purely academic. A first part of the project will be dedicated to the clinical evaluation of proposed methods on retrospective database of 250 patient through the proposed EXPERT platform. In a second part of the project, the EXPERT methodology will be evaluated in a prospective cohort in its capacity to assist the clinician in the decision making process to indicate or not a CRT implantation to any individual patient. In this observational analysis, the decision, to implant a patient or not, will only be based on guidelines and on clinician decision. The multivariate score will be evaluated, through the DSS, after pre-implantation exam and the score relevance will be evaluated 6 month after device implantation. The EXPERT project brings together three CHUs and two universities. Five experienced partners with scientific, medical, regulatory and educational skills, are united to develop a tool for exploiting massive data in echocardiography with a high-impact medical application.