An intracranial aneurysm is an anatomical abnormality resulting from a structural and persistent deformation of the wall of a cerebral artery. It is estimated that between 2 and 5% of the world’s population has a cerebral aneurysm. The annual risk of rupture ranges from 1 to 4% among those with the condition—or 7–11 per 100,000—making it a major public health issue. Intracranial aneurysms are generally discovered only upon rupture, which results, in about one out of every four cases, in the patient’s death before even arriving at the hospital. For patients treated quickly enough, about half die within the month following hospitalization, and one in three survivors will experience significant neurological deficits. There are also cases where an aneurysm is discovered incidentally during a brain imaging exam. The clinician must then, based on the anatomical images, assess and predict the risk of potential aneurysm rupture before deciding on the most appropriate treatment (neurosurgery, endovascular treatment, or a conservative approach). The decision to intervene or not to intervene (conservative management) is based on morphological criteria (size, shape, etc.) and the aneurysm’s location, as well as on epidemiological factors such as the patient’s high blood pressure, alcohol consumption, and tobacco use. While this information allows for an initial assessment of the risk of potential aneurysm rupture, it provides no insight into the biomechanical quality of the aneurysm wall, which remains the most important factor in determining the probability of rupture. Knowledge of these biomechanical factors regarding the vulnerability of Intracranial aneurysms would be a valuable diagnostic aid for clinicians and would enable patients to receive the best possible care based on the stage of maturation of their condition.
Objectives of the Dissertation
The purpose of this dissertation is to propose, based on a) images from animal experiments and b) images obtained from aneurysm phantoms at the CNRS UMR 5205 Laboratory of Image Computing and Information Systems, INSA Lyon, Claude Bernard Lyon 1 University, Lumière Lyon 2 University, and École Centrale de Lyon, to propose an initial version of a predictive model of the mechanical behavior of the cerebral vascular wall. More specifically, the objective is to propose a model-based tool, using machine learning, that provides clinicians with quantitative information on the degree of deterioration of a patient’s vascular tissue based solely on an anatomical image.
The thesis work will focus on the following areas: Creation of a database dedicated to descriptive and predictive models:
Inventory, Preparation
The tools and models discussed in this section are data-driven, and the preparation of the data cannot be fully automated. This work therefore aims to prepare the data in a way that makes the information it contains more accessible and ensures that it describes each problem in question with all its variations; Preparation also ensures compliance with constraints in the composition of training, validation, and test sets.
Data Descriptive Models
The descriptive models proposed in this section will help organize, simplify, and facilitate understanding of the underlying information in the datasets collected and organized in the first step. This involves unsupervised learning. These models will enable work on the dataset—organized into variable instances—in which none of the explanatory variables (such as cases or case types) holds particular significance relative to the others. The proposed models will make it possible to: a) identify groups of homogeneous cases within a dataset to establish behavioral norms and, consequently, identify deviations from these norms—such as the detection of new anomalies, b) to perform data compression, c) to identify correlations between data/variables, and d) to identify the discriminating data/variables correlated
with cases of anomalies.
Predictive Model
The predictive models proposed in this third section aim to predict the observable phenomenon—aneurysm progression—as effectively described by biomechanical models
and based on images. Specifically, one or more variables—defined based on features extracted from the images and biomechanical data—will be identified as the most discriminative in assessing the probability of a particular progression of the aneurysm’s condition. Two categories of operations will be considered: discrimination or classification, and regression or prediction, depending on the type of variable to be explained. Finally, the category of asymmetric models will be explored. In this category, we have data from different modalities in the training set and at the time of testing (classification/regression). In this case, we will assume that an image is provided at the time of testing—the online stage—and that the biomechanical data from the CNRS UMR 5205 Laboratory of Image Informatics and Information Systems, INSA Lyon, Claude Bernard Lyon 1 University, Lumière Lyon 2 University, and École Centrale de Lyon—linked by a correlation established beforehand, during the offline or deferred stage—constitute the training set. This work is part of an established multidisciplinary collaborative framework between ECL and HCL