Dyspnea Facial Recognition During Weaning

Overview

Critically ill patients are exposed to many sources of discomfort and traumatic experiences, especially if they require invasive mechanical ventilation (IMV). Dyspnea, or sensation of "not getting enough air - suffocation" is the most common and distressing symptom experienced by IMV patients, far more unpleasant than pain. But, contrarily to pain, dyspnea has received only little attention and is still markedly under-recognized in IMV patients. Moreover, given the deleterious short- and long-terms consequences of letting IMV patients with dyspnea, its assessment and treatment figures among the main next great cause in critical healthcare. However, dyspnea assessment in IMV patients is a challenge since many of them cannot express their suffering (e.g. sedative drugs, mouth tubes). Dyspnea observation scales (DOS) are promising alternatives that allow to strongly suspecting dyspnea. These scales encompass the dyspnea multidimensionality assessing the respiratory drive (respiratory rate, excessive use neck muscle, nasal flaring), neurovegetative signs (heart rate) and emotions (fearful face). DOS allows calculation of scores strongly correlated with dyspnea in IMV communicative patients and responsive to dyspnea treatment even in noncommunicative ones. However these scales (1) require human resources, (2) still elicit caregivers' subjectivity (fearful face), and (3) are discontinuous, whereas dyspnea is unpredictable, and thus may lead to false appreciation of clinical deterioration. Thus there is an urgent unmet need for technology-enhanced clinical surveillance tools that reliably detect dyspnea in IMV patients and tailor its relief. Infrared thermal imaging (IRTI) offers a unique opportunity to automatically and continuously compute DOS. Indeed, it has been demonstrated as reliable to measure heart and respiratory rate in patients and detect facial expressions even during surgical intervention. The study goal is to prove the concept that IRTI camera device is feasible and reliable to strongly suspect dyspnea, based on the calculation of DOS including heart and respiratory rate, facial expression of fear, and activation of Alae nasi muscle, in IMV patients experiencing an asphyxial threat during a spontaneous breathing trial. The second study goal is to assess the performance of of this multidimensional video taped monitoring to predict the outcome of the spontaneous breathing trial. This project deals with the perspective that artificial intelligence and the development of autonomous patient-machine interfaces will give access to patients' emotions by the analysis of behaviors including facial expressions, in order to improve comfort and reduce traumatic memories of the ICU stay.

Scientific background and rational of the project Dyspnea, defined as "suffocation or not getting enough air" is the most common (50% of patients) and debilitating symptom experienced by patients receiving invasive mechanical ventilation (IMV). Dyspnea is a noxious sensation but is far more unpleasant than pain because it is constantly associated with anxiety (up to unquenchable fear of dying), and because patients cannot escape from their source of dyspnea, which also connect them to life (ventilator). The inability to verbally communicate with caregivers reinforces the sense of loss of control and fear. Apart form generating immediate suffering, dyspnea is associated with negative ventilator weaning outcomes and plays a major role in the genesis of post traumatic stress disorders which concerns almost 20% of IMV patients. This should make it a major preoccupation for ICU physicians and nurses whose mission is to relieve symptoms. However, very little attention is given to dyspnea in IMV patients, which remains markedly under-recognized. Besides this caregivers' unawareness it might be even more challenging to detect dyspnea in IMV patients, since at least one-third are unable to communicate with their caregivers (sedative drugs, delirium, endotracheal tube). However, the inability to communicate in no way negates the possibility that an individual is experiencing dyspnea. Noncommunicative IMV patients are exposed to the same risk factors of dyspnea as communicative patients and should rather be considered as a vulnerable population at high risk for misdiagnosis and under treatment of dyspnea. In view of the listed issues, the use of a medical solution to robustly suspect dyspnea without eliciting patient cooperation is of utmost importance. It will warn caregiver to the possibility of dyspnea in a patient and will prompt patient/caregiver interactions aiming at relieve this discomfort, such as adjusting ventilator settings. Dyspnea observation scales (DOS) do not require patient's self-report and are promising alternatives that overcome this "dyspnea invisibility". These DOS encompass the dyspnea multidimensionality and are based on physiological and emotional modification the most correlated to dyspnea including tachycardia (neurovegetative sign), facial expression of fear (emotional consequence) and observational signs of respiratory drive such as breathing frequency, excessive use extradiaphragmatic inspiratory muscle (neck or Alae nasi muscle). These DOS allows calculation of scores strongly correlated with dyspnea in communicative IMV patients and responsive to dyspnea treatment in noncommunicative ones. Benefits and limits of dyspnea observation scales The main advantage of these multidimensional scales is, exactly, to limit observer subjectivity where heuristic reasoning (intuition) and medical experience have failed. Such numerical transformation of clinical variables allows statistics application for clinical research and monitoring of symptoms evolution to tailor therapeutic interventions. However, DOS contain several limits in routine. Firstly, their clinical use requires human medical resources and time. Secondly, inter-observer agreement between physicians and nurses is moderate and heterogeneous among items, being at its lowest for "facial expression of fear" traducing substantial observer's subjectivity involvement that may exactly limits the specific advantages of such tools. Thirdly, these scales can, by nature, only provide discontinuous measurements" whereas dyspnea is by nature labile. As a result, the use of DOS, even if it constitutes an improvement over an absence of detection, cannot guarantee an efficient monitoring of clinical deterioration. Infrared thermography: toward Autonomous Non-contact Patient Monitoring Infrared thermal imaging (IRTI) offers a unique opportunity to automatically and continuously compute DOS. Indeed, IRTI is known to give access to respiratory rate through cyclic detection of warm expired air at the mouth/nose or endotracheal tube, heart rate through cyclic warming of the face related to the cardiac pulse, and facial expression through algorithm of facial expressions recognition according to the Facial Action Coding System (FACS). Main advantages of IRTI compared to standard Red/Green/Blue signal, is its ability to detect respiration from the face and its operation even in total darkness providing accurate monitoring of patients also during night, in the ICU setting. In other hand, current contact methods of cardiorespiratory monitoring often cause discomfort or soreness (e.g. undesirable skin irritation, wire discomfort during sleep) and contact surface electrodes detach easily from the skin, especially in dyspneic patients (agitated and sweaty) and depend on patients' morphology (e.g. obese patients or thoracic deformation). For all these reasons, IRTI appears as an unobtrusive providential solution combining all these multidimensional measurements in one single technological camera device, and restoring the possibility to continuously suspect dyspnea in IMV patients, without elicit their conscious cooperation. Preliminary data of facial expressions of dyspnea If the detection of respiratory and heart rate may appear obvious and are well validated but this is not necessarily the case for facial expression. Here, the investigators planned to focus on the detection of the FACS Action Units (AU) 1 (Inner brow raiser), 2 (outer brow raiser), 5 (Upper lid raiser) and 38 (nasal flaring) Excessive use of neck muscle is a classical "respiratory drive - item" of the DOS, but is not accessible with a face focused camera and could be replace by the cyclic contraction of the Alae Nasi muscle (nasal flaring). Indeed nasal flaring is listed as item of the classical DOS, are well known as an indicator of respiratory drive and have been shown by the investigator group to strongly correlate with both experimental and in clinical ICU dyspnea. Facial expression of fear is also listed as item of classical DOS, and is well described by the concomitant onset of AU1, 2 and 5. The investigator group has recently confirmed that AU1, 2, 5 are the facial expressions the most prevalent in experimentally induced dyspnea among 23 healthy volunteers. In this data setup, AU38 had a strong discriminant value between pain and dyspnea.