Détection automatique de déviations chirurgicales et identification de comportements chirurgicaux par modélisation et analyse des processus chirurgicaux.

Abstract : Adverse events are an important concern for medical domain, their reduction is searched to allow the best safety for patients. The adverse events are, according to the HAS (Haute Autorité de santé), ‘‘situations which divert from procedures or from expected results in a usual situation and which are or which would be potentially sources of damage’’. Even though postoperative adverse events have been studied for many years, the ones which occur during the operation are recently studied, for example the first classification of intraoperative adverse events is the classification of Kaafarani et al. published in 2014. Nevertheless, the classification of intraoperative adverse events is only the first step to understand the surgical behaviors to their sources. In this thesis, we will present methods to detect the apparition of deviations due to intraoperative adverse events and to identify surgical behaviors thanks to surgical process model. To allow the development of these methods, the first step was to model the rectopexy and the adverse events related to this surgery thanks to the creation of ontologies. This work has enabled us to understand the principle of this operation and to create a vocabulary. This vocabulary was used to annotate laparoscopic videos of rectopexies, in order to create surgical process models Thanks to the surgical video annotation based on this modelisation of the rectopexy, we have developed a method to automatically detect deviations due to adverse events. This method is based on a multidimensional non-linear temporal scaling, a homemade alignment of sequences, follows by a hidden semi-Markovian model. This Markovian model was trained to detect deviations from a standard surgical process, a reference, and to determine if these deviations are due to adverse events. This deviation detection is the first step in order to understand the reason of their apparitions. We hypothesize that their apparitions could be explained by an activities succession, i.e. a pattern. To verify this hypothesis, we develop a pattern discovery method to allow the identification of specific surgical behaviors. This identification of surgical behaviors was done by a hierarchical clustering thanks to a new metric based on shared pattern between surgeries. To validate our method, we make a comparison with two state of the art article highlighting surgical behaviors, for example, surgical behaviors specific to surgical site or to type of procedures. Once our method has been validated, we have used it to identify surgical behavior specific to preoperative data and to adverse events apparitions. Finally, we come back to the most important contributions of this work through a general discussion and we propose perspectives to improve our results.
Complete list of metadatas

Cited literature [124 references]  Display  Hide  Download

Contributor : Arnaud Huaulmé <>
Submitted on : Friday, May 5, 2017 - 12:00:25 PM
Last modification on : Thursday, March 21, 2019 - 2:56:08 PM
Long-term archiving on : Sunday, August 6, 2017 - 12:34:06 PM


  • HAL Id : tel-01518734, version 1


Arnaud Huaulmé. Détection automatique de déviations chirurgicales et identification de comportements chirurgicaux par modélisation et analyse des processus chirurgicaux.. Traitement du signal et de l'image [eess.SP]. Université Grenoble - Alpes, 2017. Français. ⟨tel-01518734⟩



Record views


Files downloads