Project

Back to overview

Model observers for detection in 3D medical imaging

English title Model observers for detection in 3D medical imaging
Applicant Bochud François
Number 156032
Funding scheme Project funding (Div. I-III)
Research institution Institut de Radiophysique Département de Radiologie Université de Lausanne/CHUV
Institution of higher education University of Lausanne - LA
Main discipline Cancer
Start/End 01.01.2015 - 31.12.2017
Approved amount 195'798.00
Show all

All Disciplines (2)

Discipline
Cancer
Other disciplines of Physics

Keywords (3)

detection; computed tomography; model observer

Lay Summary (French)

Lead
Ces dernières décennies, l'imagerie tridimensionnelle s'est généralisée et les doses délivrées par les examens de radiodiagnostic ont augmenté régulièrement. La dose ne pouvant être efficacement optimisée que si la qualité d'image est objectivement quantifiée, nous proposons d'utiliser des modèles d'observateur tridimensionnels afin d'estimer la performance d'un observateur humain dont la tâche consiste à détecter une pathologie dans une image de scanner RX.
Lay summary

Contenu et objectifs du travail de recherche

Nous proposons de développer deux types de modèles d'observateur.  Le premier sera une extension directe de ce qui s'est déjà fait à deux dimensions et incorporera l'effet du défilement au travers des coupes. Le second modèle intégrera une composante de recherche qui stimulera ce qui se passe dans la périphérie de la vision.  Le but ultime est de pouvoir estimer la qualité d'image dans des cas proches de la clinique.

Ces modèles seront testés sur des images de tomosynthèse mammaire et des images de scanner RX du foie obtenues au département de radiologie du CHUV, Lausanne.  Leurs performances seront comparées à celles d'observateurs humains afin d'ajuster et de valider leurs caractéristiques.  De manière ultime, nous désirons proposer un fantôme imprimé en 3D qui puisse être utilisé pour estimer la performance des modèles.

 

Contexte scientifique et social du projet de recherche

Contrairement aux modèles d'observateur actuels, nous proposons de prendre en compte deux aspects importants de la vision humaine dans ce travail : la capacité que nous avons à scruter une image à trois dimension et à localiser un signal.  Ceci devrait nous permettre de définir une qualité d'image de manière plus proche de la pratique clinique et ainsi d'améliorer la manière dont la communauté médicale optimise la dose au patient en radiologie.
Direct link to Lay Summary Last update: 13.10.2014

Responsible applicant and co-applicants

Employees

Name Institute

Publications

Publication
Channelized Hotelling observer correlation with human observers for low-contrast detection in liver CT images
Ba Alexandre, Abbey Craig K., Racine Damien, Viry Anaïs, Verdun Francis R., Schmidt Sabine, Bochud François O. (2019), Channelized Hotelling observer correlation with human observers for low-contrast detection in liver CT images, in Journal of Medical Imaging, 6(02), 1-1.
Inter‐laboratory comparison of channelized hotelling observer computation
Ba Alexandre, Abbey Craig K., Baek Jongduk, Han Minah, Bouwman Ramona W., Balta Christiana, Brankov Jovan, Massanes Francesc, Gifford Howard C., Hernandez‐Giron Irene, Veldkamp Wouter J. H., Petrov Dimitar, Marshall Nicholas, Samuelson Frank W., Zeng Rongping, Solomon Justin B., Samei Ehsan, Timberg Pontus, Förnvik Hannie, Reiser Ingrid, Yu Lifeng, Gong Hao, Bochud François O. (2018), Inter‐laboratory comparison of channelized hotelling observer computation, in Medical Physics, 45(7), 3019-3030.
Low contrast detection in abdominal CT: comparing single-slice and multi-slice tasks
Ba Alexandre, Racine Damien, Viry Anaïs, Verdun Francis R., Schmidt Sabine (2017), Low contrast detection in abdominal CT: comparing single-slice and multi-slice tasks, in SPIE Medical Imaging, Orlando, Florida, United StatesSociety of Photo-Optical Instrumentation Engineers (SPIE)., Bellingham, WA USA.
Anthropomorphic model observer performance in three-dimensional detection task for low-contrast computed tomography
Ba Alexandre, Eckstein Miguel P., Racine Damien, Ott Julien G., Verdun Francis, Kobbe-Schmidt Sabine, Bochud François O. (2016), Anthropomorphic model observer performance in three-dimensional detection task for low-contrast computed tomography, in Journal of Medical Imaging, (1), 011009-011009.
Low contrast detectability in CT for human and model observers in multi-slice data sets
Ba Alexandre, Racine Damien, Ott Julien G., Verdun Francis R., Kobbe-Schmidt Sabine, Eckstein Miguel P., Bochud Francois O. (2015), Low contrast detectability in CT for human and model observers in multi-slice data sets, in SPIE Medical Imaging, 9.

Collaboration

Group / person Country
Types of collaboration
Univ. California, Santa Barbara United States of America (North America)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure

Scientific events

Active participation

Title Type of contribution Title of article or contribution Date Place Persons involved
SPIE Medical Imaging Talk given at a conference Low contrast detection in abdominal CT: comparing single-slice and multi-slice tasks 13.02.2017 Orlando, United States of America Ba Alexandre;
SPIE Medical Imaging Talk given at a conference Low contrast detectability in CT for human and model observers in multi-slice data sets 16.02.2015 Orlando, United States of America Ba Alexandre;


Communication with the public

Communication Title Media Place Year
Talks/events/exhibitions Défense publique de thèse : Les modèles d'observateur pour l'évaluation de la qualité de l'image Western Switzerland 2018

Associated projects

Number Title Start Funding scheme
197477 Medical image quality: Model observers of radiologists performing a search task and AI algorithms that include the anatomical texture in low-contrast CT diagnostics 01.02.2021 Project funding (Div. I-III)
135668 3D model observer for detection in CT imaging 01.05.2011 Project funding (Div. I-III)

Abstract

Background. Radiation dose to the patient in medical diagnostic imaging has been steadily increasing together with the use of 3D modalities for the last two decades. Dose can only be optimized with the knowledge of a pertinent estimation of image quality. Mathematical model observers have already shown their usefulness for 2D image modalities, but their generalization into 3D modalities is still in the limbo. In this work, we propose to develop two anthropomorphic model observers able to estimate the performance of human observers searching for and detecting a pathology in 3D computed tomography.Aims and objectives. We propose to develop model observers. The first will be a direct extension of conventional detection 2D detection model observers that will incorporate the scrolling action within the image stack. The second model will incorporate a search component that simulates what is happening in the vision periphery. The ultimate goal is to provide a practical method to measure image quality in the clinical practice.Material and methods. The first model (TRF-CHO) will be a channelized Hotelling observer (CHO) able to detect a signal at a known location on a 3D image stack. The effect of navigating through the image stack will be taken into account by a time response function (TRF) developed in a previous project. The second model observer, called "search extra-foveal observer" (SEFO), will be developed together with our colleagues from the University of California, Santa Barbara (Dr Eckstein's lab). It builds on recent advances in neuroscience and processes the visual field with decreasing spatial resolution with increasing distance from the point of fixation. Both models will be tested on digital breast tomosynthesis and hepatic nodule CT exams provided by the department of radiology at CHUV. Their performances will be compared with those of human observers in order to validate them and tune their characteristics (our psychophysics laboratory is equipped with an infra-red video eye-tracker). Ultimately, we want to propose a 3D printed phantom with anatomic-like structures that could be imaged in order to compute the performance of the model observers. The 3D printed phantoms will be produced by our colleagues of Duke University (Dr Samei's lab).Expected results and importance. Contrary to present model observers we will incorporate two key aspects of human vision: the ability to scan an image stack and to search a signal location. This should provide an image quality parameter that is closer to the common practice than the present situation. This should be welcome by the medical community in charge of optimizing radiation dose.
-