Ph. D. Project
Title:
Development of a tool to assist awake surgery of brain tumors based on statistical learning methods
Dates:
2024/10/01 - 2027/09/30
Supervisor(s):
Other supervisor(s):
Dr MEZIERES Sophie (sophie.mezieres@univ-lorraine.fr)
Description:
Context
Diffuse low grade gliomas are infiltrating brain tumors whose evolution is systematically towards anaplastic transformation. The therapeutic strategy consists in reducing the tumor volume to limit this risk of transformation and increase survival [1]. Surgery is the first-line treatment. Since these tumors infiltrate the functional structures of the brain, as we have already demonstrated [2], surgery must be performed in awake condition, with active participation of the patient while electrical brain stimulation is performed to identify functional structures. These awake surgeries have thus shown a benefit in terms of quality of removal and preservation of brain functions.
However, progressive variations in patient performance can be observed during surgery, as well as some fine disorders (e.g.: time to complete the test), difficult to apprehend by the evaluator but with possible postoperative consequences.
In Nancy, these surgeries are filmed in order to be able to return to possible disorders per or postoperative and understand the role of stimulated brain structures. This video data will be used in the framework of this thesis.
Subject Description
The main objective of this topic is to develop a tool to assist awake surgery based on statistical learning methods. This tool should make it possible to better identify fine disorders during surgery, impossible or difficult to assess by humans, in order to better organize the gesture (detection of possible "tipping points" beyond which the patient is no longer performing) or to stop it (detection of generally minimal alterations but which in the socio-professional context of the patient will have deleterious consequences on his quality of life). A secondary objective is the realization, based on the same methods, of a tool allowing to detect the operating phases in order to realize adapted structure-function correlations (detection of the stimulated area on the video and the behavioral anomaly with recording).
The two main scientific questions we wish to address in this thesis are:
How to automatically detect, analyze and record the patient's responses (motor skills, speech) to different stimuli during surgery?
Do the fine abnormalities identified allow for the extraction of predictors of short- and long-term neurocognitive status?
Regarding the first question, the current procedure of awake surgery is not automated. Note-taking of events related to stimulation is manual, therefore imprecise, or even impossible for certain fine parameters, including chronometry The interpretation of these events during surgery and during post-surgery follow-up is both very important and time-consuming. Despite available video data, it remains difficult to achieve given the length of the awake phase to be analyzed (2 to 3 hours). Thanks to the contribution of artificial intelligence, and in particular deep learning, we propose to automate part of the procedure of analysis of responses to stimulation, through motion recognition and speech recognition algorithms that we will transpose to the conditions of awake surgery. This approach will allow us to measure new parameters hitherto unexplored, for example variations in the speed of movements during the intervention or response times... It will also identify in an off-line video the key moments concerning a function or a structure to optimize the procedure later and to establish anatomo-functional correlations.
The acquisition of these data and their precise identification should make it possible to answer the second question by correlating these intraoperative data to the short and long term cognitive future of patients. This return could finally allow to define intraoperative indicators of long-term cognitive risk. As this issue has not yet been addressed, it is exploratory but of major interest for the quality of life of patients at the end of the intervention. Such indicators, if sufficiently robust, could make it possible to define an intraoperative model predictive of the risk of long-term cognitive deterioration on functions difficult to test such as working memory, flexibility, attention or concentration for example.
The work on the two scientific questions set out should lead to the development of a tool to assist the surgeon in awake surgeries and a better knowledge of brain functioning.
Positioning
Currently awake surgeries are performed to varying degrees in most neurosurgery centers. The activity in Nancy is particularly important because of recruitment in the region and the scientific level concerning this theme, as well as because of historical collaborations with the international reference center of the University Hospital of Montpellier (team of Professor Hugues Duffau). In most of the centers that carry out important, as is also the case at the Lariboisière Hospital in Paris (team of Professor Emmanuel Mandonnet), they are recorded but the video data are difficult to use because of the lack of tools to identify important phases automatically. The objective of this work is to overcome this major difficulty.
Work requested
The work requested will include a state of the art on:
- low grade gliomas and awake surgery;
- deep learning AI techniques;
- Statistical learning methods (unsupervised and supervised);
This will be followed by the development of models adapted to the constraints of awake surgery to feed deep learning algorithms in order to build a tool for automatic recognition of movement and speech.
It will then attempt to correlate the variables acquired to postoperative outcomes in the short and long-term using pre- and postoperative cognitive assessments.
Diffuse low grade gliomas are infiltrating brain tumors whose evolution is systematically towards anaplastic transformation. The therapeutic strategy consists in reducing the tumor volume to limit this risk of transformation and increase survival [1]. Surgery is the first-line treatment. Since these tumors infiltrate the functional structures of the brain, as we have already demonstrated [2], surgery must be performed in awake condition, with active participation of the patient while electrical brain stimulation is performed to identify functional structures. These awake surgeries have thus shown a benefit in terms of quality of removal and preservation of brain functions.
However, progressive variations in patient performance can be observed during surgery, as well as some fine disorders (e.g.: time to complete the test), difficult to apprehend by the evaluator but with possible postoperative consequences.
In Nancy, these surgeries are filmed in order to be able to return to possible disorders per or postoperative and understand the role of stimulated brain structures. This video data will be used in the framework of this thesis.
Subject Description
The main objective of this topic is to develop a tool to assist awake surgery based on statistical learning methods. This tool should make it possible to better identify fine disorders during surgery, impossible or difficult to assess by humans, in order to better organize the gesture (detection of possible "tipping points" beyond which the patient is no longer performing) or to stop it (detection of generally minimal alterations but which in the socio-professional context of the patient will have deleterious consequences on his quality of life). A secondary objective is the realization, based on the same methods, of a tool allowing to detect the operating phases in order to realize adapted structure-function correlations (detection of the stimulated area on the video and the behavioral anomaly with recording).
The two main scientific questions we wish to address in this thesis are:
How to automatically detect, analyze and record the patient's responses (motor skills, speech) to different stimuli during surgery?
Do the fine abnormalities identified allow for the extraction of predictors of short- and long-term neurocognitive status?
Regarding the first question, the current procedure of awake surgery is not automated. Note-taking of events related to stimulation is manual, therefore imprecise, or even impossible for certain fine parameters, including chronometry The interpretation of these events during surgery and during post-surgery follow-up is both very important and time-consuming. Despite available video data, it remains difficult to achieve given the length of the awake phase to be analyzed (2 to 3 hours). Thanks to the contribution of artificial intelligence, and in particular deep learning, we propose to automate part of the procedure of analysis of responses to stimulation, through motion recognition and speech recognition algorithms that we will transpose to the conditions of awake surgery. This approach will allow us to measure new parameters hitherto unexplored, for example variations in the speed of movements during the intervention or response times... It will also identify in an off-line video the key moments concerning a function or a structure to optimize the procedure later and to establish anatomo-functional correlations.
The acquisition of these data and their precise identification should make it possible to answer the second question by correlating these intraoperative data to the short and long term cognitive future of patients. This return could finally allow to define intraoperative indicators of long-term cognitive risk. As this issue has not yet been addressed, it is exploratory but of major interest for the quality of life of patients at the end of the intervention. Such indicators, if sufficiently robust, could make it possible to define an intraoperative model predictive of the risk of long-term cognitive deterioration on functions difficult to test such as working memory, flexibility, attention or concentration for example.
The work on the two scientific questions set out should lead to the development of a tool to assist the surgeon in awake surgeries and a better knowledge of brain functioning.
Positioning
Currently awake surgeries are performed to varying degrees in most neurosurgery centers. The activity in Nancy is particularly important because of recruitment in the region and the scientific level concerning this theme, as well as because of historical collaborations with the international reference center of the University Hospital of Montpellier (team of Professor Hugues Duffau). In most of the centers that carry out important, as is also the case at the Lariboisière Hospital in Paris (team of Professor Emmanuel Mandonnet), they are recorded but the video data are difficult to use because of the lack of tools to identify important phases automatically. The objective of this work is to overcome this major difficulty.
Work requested
The work requested will include a state of the art on:
- low grade gliomas and awake surgery;
- deep learning AI techniques;
- Statistical learning methods (unsupervised and supervised);
This will be followed by the development of models adapted to the constraints of awake surgery to feed deep learning algorithms in order to build a tool for automatic recognition of movement and speech.
It will then attempt to correlate the variables acquired to postoperative outcomes in the short and long-term using pre- and postoperative cognitive assessments.
Keywords:
deep learning, modeling, image and video processing, statistical learning
Department(s):
| Biology, Signals and Systems in Cancer and Neuroscience |
Publications: