300 research outputs found
Personalization of Reaction-Diffusion Tumor Growth Models in MR Images: Application to Brain Gliomas Characterization and Radiotherapy Planning
No full textInternational audienceReaction-diffusion based tumor growth models have been widely used in the literature for modeling the growth of brain gliomas. Lately, recent models have started integrating medical images, specifically anatomical and diffusion images, in their formulation. On the other hand, the adaptation of the general model to the specific patient cases has not been studied thoroughly yet. In this chapter we address this adaptation. This chapter is a short summary of the articles (Konukoglu 2009a), (Konukoglu 2009b) and the thesis (Konukoglu 2009c) that we have submitted recently. In the first part, we describe a parameter estimation method for reaction-diffusion tumor growth models using time series of medical (Magnetic Resonance) images. This method estimates the patient specific parameters of the model using the images of the patient taken at different successive time instances. In the second part of the chapter we focus on an application of the personalized models aimed to improve the tumor targeting in radiation therapy. Specifically we address the problem of limited visualization of medical images. We describe a method for extrapolating the invisible infiltration margins of gliomas in the MR images and the usage of these margins in constructing irradiation margins taking into account the growth dynamics of the tumor. Finally for both parts we show preliminary results demonstrating the power and the potential benefits of the personalizatio
Vision-Based Neurosurgical Guidance: Unsupervised Localization and Camera-Pose Prediction
No full textLocalizing oneself during endoscopic procedures can be problematic
due to the lack of distinguishable textures and landmarks, as well
as difficulties due to the endoscopic device such as a limited field of view
and challenging lighting conditions. Expert knowledge shaped by years
of experience is required for localization within the human body during
endoscopic procedures. In this work, we present a deep learning method
based on anatomy recognition, that constructs a surgical path in an unsupervised
manner from surgical videos, modelling relative location and
variations due to different viewing angles. At inference time, the model
can map an unseen video’s frames on the path and estimate the viewing
angle, aiming to provide guidance, for instance, to reach a particular destination.
We test the method on a dataset consisting of surgical videos of
transsphenoidal adenomectomies, as well as on a synthetic dataset. An
online tool that lets researchers upload their surgical videos to obtain
anatomy detections and the weights of the trained YOLOv7 model are
available at: https://surgicalvision.bmic.ethz.ch
Navigating the Unknowns of Medical Imaging
No full textThe integration of Artificial Intelligence (AI) in medical imaging has the potential of revolutionizing healthcare, allowing professionals to analyze the details of the human body with unprecedented accuracy and speed. Despite the potential of AI to transform medical diagnostics, it faces significant challenges, primarily due to limited data sets and the difficulty in generalizing across diverse medical scenarios. Large datasets like the UK Biobank and TCGA provide a foundation, but generalization and reliability in diverse situations remain hurdles. Particularly challenging are Out-of-Distribution (OOD) shifts caused by demographic changes, advancements in imaging technologies, and evolution in clinical practices, which pose risks to the reliability and trustworthiness of AI systems. One of the critical issues is the variability in cancer subtypes, which, despite similar visual presentations, vary widely in prognosis. Training models to accurately differentiate these subtypes is hindered by data privacy laws and the scarcity of samples for rare subtypes. The COVID-19 pandemic further exemplified the shortcomings of current state-of-art in medical imaging AI. These scenarios underscore the need for AI systems that can adapt swiftly to new challenges, maintaining reliable support in critical diagnostic processes. This dissertation addresses these challenges by proposing methodologies to enhance the reliability of Deep Learning (DL)-based systems in medical imaging. Key approaches include Multi-headed Varational Inference (VIMH) for uncerainty estimation, Sliding-Window Optimal Transport for OOD Detection (SWOT), and Histopathology Artifact Restoration Pipeline (HARP). While VIMH offers improved robustness against OOD shifts, it demands specialized training. In contrast, SWOT and HARP offer post-hoc solutions applicable to existing AI models, enhancing diagnostic precision and ensuring reliability in the deployed AI systems. The dissertation also explores dynamic learning settings such as Continual Learning (CL) and Federated Learning (FL). In the federated histopathology settings, Federated Stain Normalization with BottleGAN (BottleGAN) presents an ideal solution for overcoming limited data annotation and data heterogeneity in Computational Pathology (CP). Furthermore, Closing-the-Loop with Radiologists (CtLwR) integrates transparency into the AI decision-making process and leverages structured report to enable CL. These approaches are vital in adapting to varying populations and institutional changes, maintaining the reliability of DL-based models. In conclusion, the medical imaging sector’s increasing use of AI requires a balanced approach that prioritizes patient privacy, reliability, and ethical standards. The shift from AI vs. clinicians to AI collaborating with clinicians signals a significant change, combining efficiency with patient-focused care. To successfully navigate the unknowns of the medical field, AI must address the challenges and ensure its integration into healthcare is beneficial and safe
Precision Medicine in Head-and-Neck Cancer: Utilizing Artificial Intelligence to Improve Diagnosis and Treatment
No full textModélisation de la croissance des gliomes et personnalisation des modéles de croissance à l'aide d'images médicales
No full textMathematical models and more specifically reaction-diffusion based models have been widely used in the literature for modeling the growth of brain gliomas and tumors in general. Besides the vast amount of research focused on microscopic and biological experiments, recently models have started integrating medical images in their formulations. By including the geometry of the brain and the tumor, the different tissue structures and the diffusion images, models are able to simulate the macroscopic growth observable in the images. Although generic models have been proposed, methods for adapting these models to individual patient images remain an unexplored area. In this thesis we address the problem of "personalizing mathematical tumor growth models". We focus on reaction-diffusion models and their applications on modeling the growth of brain gliomas. As a first step, we propose a method for automatic identification of patient-specific model parameters from series of medical images. Observing the discrepancies between the visualization of gliomas in MR images and the reaction-diffusion models, we derive a novel formulation for explaining the evolution of the tumor delineation. This "modified anisotropic Eikonal model" is later used for estimating the model parameters from images. Thorough analysis on synthetic dataset validates the proposed method theoretically and also gives us insights on the nature of the underlying problem. Preliminary results on real cases show promising potentials of the parameter estimation method and the reaction-diffusion models both for quantifying tumor growth and also for predicting future evolution of the pathology. Following the personalization, we focus on the clinical application of such patient-specific models. Specifically, we tackle the problem of limited visualization of glioma infiltration in MR images. The images only show a part of the tumor and mask the low density invasion. This missing information is crucial for radiotherapy and other types of treatment. We propose a formulation for this problem based on the patient-specific models. In the analysis we also show the potential benefits of such the proposed method for radiotherapy planning. The last part of this thesis deals with numerical methods for anisotropic Eikonal equations. This type of equation arises in both of the previous parts of this thesis. Moreover, such equations are also used in different modeling problems, computer vision, geometrical optics and other different fields. We propose a numerical method for solving anisotropic Eikonal equations in a fast and accurate manner. By comparing it with a state-of-the-art method we demonstrate the advantages of our technique.Les modèles mathématiques et plus spécifiquement les modèles basés sur l'équation de réaction-diffusion ont été utilisés largement dans la littérature pour modéliser la croissance des gliomes cérébraux et des tumeurs en général. De plus la grande littérature de recherche qui concentre sur les expériences biologiques et microscopiques, récemment les modèles ont commencé intégrer l'imagerie médicale dans ses formulations. Incluant la géométrie du cerveau et celle de la tumeur, les structures des différentes tissues et la direction de diffusion, ils ont montré qu'il est possible de simuler la croissance de la tumeur comme c'est observé dans les images médicales. Bien que des modèles génériques ont été proposés, les méthodes pour adapter ces modèles aux images d'un patient reste un domaine inexploré. Dans cette thèse nous nous adressons au problème de 'personnalisation de modèle mathématique de la croissance de tumeurs'. Nous nous focalisons sur les modèles de réaction-diffusion et leurs applications sur la croissance des gliomes cérébrales. Dans la première étape, nous proposons une méthode pour l'identification automatique des paramètres 'patient-spécifiques' du modèle à partir d'une série d'images. En observant la divergence entre la visualisation des gliomes dans les IRMs et les modèles réaction-diffusion, nous déduisons une nouvelle formulation pour expliquer l'évolution de la délinéation de la tumeur. Ce modèle 'Eikonal anistropique modifié' est utilisé plus tard pour l'estimation des paramètres à partir des images. Nous avons théoriquement analysé la méthode proposée à l'aide d'un base donne synthétique et nous avons montré la capacité de la méthode et aussi sa limitation. En plus, les résultats préliminaires, sur les cas réels montrent des potentiels prometteurs de la méthode d'estimation des paramètres et du modèle de réaction-diffusion pour la quantification de la croissance de tumeur et aussi pour la prédiction de l'évolution futur de la tumeur. En suivant la personnalisation, nous nous concentrons sur les applications cliniques des modèles 'patient-spécifiques'. Spécifiquement, nous nous attaquons au problème de la visualisation limitée d'infiltration de gliome dans l'IRM. En effet, les images ne montrent qu'une partie de la tumeur et masquent l'infiltration basse-densité. Cette information absente est cruciale pour la radiothérapie et aussi pour d'autre type de traitements. Dans ce travail, nous proposons pour ce problème une formulation basée sur les modèles 'patient-spécifiques'. Dans l'analyse de cette méthode nous montrons également les bénéfices potentiels pour la planification de la radiothérapie. La dernière étape de cette thèse se concentre sur les méthodes numériques de l'équation 'Eikonal anisotropique'. Ce type d'équation est utilisé dans beaucoup de problèmes différents tel que la modélisation, le traitement d'image, la vision par ordinateur et l'optique géométrique. Ici nous proposons une méthode numérique rapide et efficace pour résoudre l'équation Eikonal anisotropique. En la comparant avec une autre méthode état-de-l'art nous démontrons les avantages de la technique proposée
- …
