Search of low contrast liver lesions in abdominal CT: an eye-­‐tracking study in volumetric images

Détails

Ressource 1Télécharger: Mémoire no 5528 Mme Shams.pdf (1384.86 [Ko])
Etat: Public
Version: Après imprimatur
Licence: Non spécifiée
ID Serval
serval:BIB_31A6F9BFE568
Type
Mémoire
Sous-type
(Mémoire de) maîtrise (master)
Collection
Publications
Institution
Titre
Search of low contrast liver lesions in abdominal CT: an eye-­‐tracking study in volumetric images
Auteur(s)
SHAMS M.
Directeur(s)
BOCHUD F.
Détails de l'institution
Université de Lausanne, Faculté de biologie et médecine
Statut éditorial
Acceptée
Date de publication
2018
Langue
anglais
Nombre de pages
12
Résumé
Eye-­‐tracking studies can be particularly effective in improving tumor detection by radiologists. Several studies have attempted to characterize the ability of radiologists to search for and recognize different targets in various imaging modalities. However, few studies have associated eye-­‐tracking experiments with scrolling volumetric images such as CT. Among them, a recent study on the reading of chest CT images showed that the detection strategies of radiologists could be classified in two categories, the "drillers" and the "scanners", according to an eye movement index (EMI), which quantifies the tendency of radiologists to perform large saccades in the investigated organ. However, the EMI doesn’t take into account how radiologists scroll through the different volumetric data slices. We propose to add this information through the "number of courses", defined as the number of times a reader scrolls in a given direction during the analysis of the image. Our study aims to document this quantity and show how it could complement the EMI in order to quantify the strategy of the radiologist.
We considered a set of 15 asymptomatic liver CT images in which we inserted 1 to 5 metastases of two different contrast amplitudes. Twenty radiologists were asked to search for the metastases while their eye-­‐gaze was followed by an eye-­‐tracker.
The drillers are defined a going back and forth through the image stack, each time to exploring a different area in each image. We identified them as having a low EMI (e) and a large number of courses (C). The scanners are defined as scrolling coherently through the stack of images and exploring each image slice one after the other. They tend to have a high EMI (E) and a low number of courses (c). Interestingly, we observed that radiologists with a larger number of courses (eC and EC) tended to cover more volume in more time than radiologists with a lower number of courses (ec and ec). They found more metastasis and made less search errors than those with lower number of courses, especially when searching for lower contrast signals. Therefore, a driller defined by a low EMI and a high number of courses (eC) tend to be more efficient than scanners.
Our results show that for when the task becomes more difficult, the radiologists can improve their effectiveness by applying a strategy of a driller defined as an EMI and a higher number of courses. This could be used teaching resident radiologists.
Création de la notice
03/09/2019 8:04
Dernière modification de la notice
08/09/2020 6:08
Données d'usage