Image augmentée et intelligence artificielle d’un compte rendu opératoire à la chirurgie guidée par l’ordinateur
Seance of wednesday 12 october 2022 (Que savons-nous de l’endométriose ? Quelle place pour la chirurgie)
DOI number : 10.26299/tt25-nk18/emem.2022.29.05
Abstract
L’utilisation d’aide à la chirurgie faisant appel aux technologies d’intelligence artificielle (IA) et de réalité augmentée (RA) sont porteurs de nombreuses promesses. Néanmoins le chemin reste long pour que ces technologies puissent entrer en pratique courante au bloc opératoire. Le principal problème notamment pour les technologies de machine learning (ML) reste la collecte de données. En chirurgie de l’endométriose la problématique des données est encore plus forte : les données (collecte, recueil, traitement, interprétation) cliniques, chirurgicales (vidéos notamment) ne font pas l’objet de standardisation qui seules permettent la création d’outils basée sur l’IA et la RA. Actuellement la collecte des données cliniques chirurgicales (comme Epithor ou Epicard) dans l’endométriose n’existe pas. Nous avons proposé des solutions de recueil (Application Noendo) mais qui nécessite des moyens considérables pour être déployées. Il existe d’autres applications digitales mais qui ne sont pas orientées vers le recueil des données chirurgicales. L’autre type de données chirurgicale dont on dispose est constitué des vidéos de chirurgie. A partir de ces vidéos qu’il faut prétraiter (anonymisation), on peut constituer des bases de données qui vont être annotées et permettre littéralement d’apprendre à l’ordinateur les bases de la chirurgie en commençant par un apprentissage sémiologique. Cet apprentissage est en fait un challenge majeur (collecte des données, standardisation, interprétation, création d’IA). Seule ces technologies peuvent permettre l’utilisation de la réalité augmentée en chirurgie : il faut améliorer l’autonomie d’interprétation de l’ordinateur pour qu’il puisse afficher en temps réel et au bon endroit (sur les éléments anatomiques à opérer) les données dont a besoin le chirurgien pour améliorer ses performances. Actuellement on peut afficher en réalité augmentée la structure interne de l’utérus et donc afficher l’adénomyose intra-utérine (diagnostiquée sur l’IRM) mais l’aide à la stadification, le guidage pour trouver le bon plan de dissection sont des fonctionnalités extrêmement difficiles à développer. C’est l’objectif du projet Européen FEMALE (Machine Learning en Endométriose , H2020, Financé à hauteur de 6M d’euro, dont leader est l’université d’Aarhus). Tous ces technologies doivent être basées des données solides (« smart data ») qui en font un enjeu majeur dans le développement des nouvelles technologies de RA et d’IA au bloc opératoire.
The use of surgical aids based on artificial intelligence (AI) and augmented reality (AR) technologies holds great promise. However, there is still a long way to go before these technologies can be used in the operating room. The main problem, especially for machine learning (ML) technologies, is data collection. In endometriosis surgery, the problem of data is even greater: clinical and surgical data (collection, processing, interpretation) (especially videos) are not standardized, which is the only way to create tools based on AI and AR. Currently, the collection of clinical surgical data (such as Epithor or Epicard) in endometriosis does not exist. We have proposed collection solutions (Noendo application) but they require considerable resources to be deployed. There are other digital applications but they are not oriented towards the collection of surgical data. The other type of surgical data available is surgical videos. From these videos that need to be pre-processed (anonymization), we can build databases that will be annotated and literally allow the computer to learn the basics of surgery, starting with a semiological learning. This learning is in fact a major challenge (data collection, standardization, interpretation, creation of AI). Only these technologies can allow the use of augmented reality in surgery: it is necessary to improve the autonomy of interpretation of the computer so that it can display in real time and in the right place (on the anatomical elements to be operated) the data needed by the surgeon to improve his performance. Currently, we can display in augmented reality the internal structure of the uterus and thus display intrauterine adenomyosis (diagnosed on MRI) but the assistance in staging, the guidance to find the right dissection plan are features that are extremely difficult to develop. This is the objective of the European project FEMALE (Machine Learning in Endometriosis, H2020, funded with 6M euro, led by the University of Aarhus). All these technologies must be based on solid data ("smart data") which makes them a major issue in the development of new AR and AI technologies in the operating room.
The use of surgical aids based on artificial intelligence (AI) and augmented reality (AR) technologies holds great promise. However, there is still a long way to go before these technologies can be used in the operating room. The main problem, especially for machine learning (ML) technologies, is data collection. In endometriosis surgery, the problem of data is even greater: clinical and surgical data (collection, processing, interpretation) (especially videos) are not standardized, which is the only way to create tools based on AI and AR. Currently, the collection of clinical surgical data (such as Epithor or Epicard) in endometriosis does not exist. We have proposed collection solutions (Noendo application) but they require considerable resources to be deployed. There are other digital applications but they are not oriented towards the collection of surgical data. The other type of surgical data available is surgical videos. From these videos that need to be pre-processed (anonymization), we can build databases that will be annotated and literally allow the computer to learn the basics of surgery, starting with a semiological learning. This learning is in fact a major challenge (data collection, standardization, interpretation, creation of AI). Only these technologies can allow the use of augmented reality in surgery: it is necessary to improve the autonomy of interpretation of the computer so that it can display in real time and in the right place (on the anatomical elements to be operated) the data needed by the surgeon to improve his performance. Currently, we can display in augmented reality the internal structure of the uterus and thus display intrauterine adenomyosis (diagnosed on MRI) but the assistance in staging, the guidance to find the right dissection plan are features that are extremely difficult to develop. This is the objective of the European project FEMALE (Machine Learning in Endometriosis, H2020, funded with 6M euro, led by the University of Aarhus). All these technologies must be based on solid data ("smart data") which makes them a major issue in the development of new AR and AI technologies in the operating room.