|
|
Brain Tumor Detection and Segmentation in Multisequence MRI
Dvořák, Pavel ; Přibil, Jiří (oponent) ; Šprláková-Puková,, Andrea (oponent) ; Smékal, Zdeněk (vedoucí práce)
This work deals with the brain tumor detection and segmentation in multisequence MR images with particular focus on high- and low-grade gliomas. Three methods are propose for this purpose. The first method deals with the presence detection of brain tumor structures in axial and coronal slices. This method is based on multi-resolution symmetry analysis and it was tested for T1, T2, T1C and FLAIR images. The second method deals with extraction of the whole brain tumor region, including tumor core and edema, in FLAIR and T2 images and is suitable to extract the whole brain tumor region from both 2D and 3D. It also uses the symmetry analysis approach which is followed by automatic determination of the intensity threshold from the most asymmetric parts. The third method is based on local structure prediction and it is able to segment the whole tumor region as well as tumor core and active tumor. This method takes the advantage of a fact that most medical images feature a high similarity in intensities of nearby pixels and a strong correlation of intensity profiles across different image modalities. One way of dealing with -- and even exploiting -- this correlation is the use of local image patches. In the same way, there is a high correlation between nearby labels in image annotation, a feature that has been used in the ``local structure prediction'' of local label patches. Convolutional neural network is chosen as a learning algorithm, as it is known to be suited for dealing with correlation between features. All three methods were evaluated on a public data set of 254 multisequence MR volumes being able to reach comparable results to state-of-the-art methods in much shorter computing time (order of seconds running on CPU) providing means, for example, to do online updates when aiming at an interactive segmentation.
|
|
|
Brain Tumor Detection and Segmentation in Multisequence MRI
Dvořák, Pavel ; Přibil, Jiří (oponent) ; Šprláková-Puková,, Andrea (oponent) ; Smékal, Zdeněk (vedoucí práce)
This work deals with the brain tumor detection and segmentation in multisequence MR images with particular focus on high- and low-grade gliomas. Three methods are propose for this purpose. The first method deals with the presence detection of brain tumor structures in axial and coronal slices. This method is based on multi-resolution symmetry analysis and it was tested for T1, T2, T1C and FLAIR images. The second method deals with extraction of the whole brain tumor region, including tumor core and edema, in FLAIR and T2 images and is suitable to extract the whole brain tumor region from both 2D and 3D. It also uses the symmetry analysis approach which is followed by automatic determination of the intensity threshold from the most asymmetric parts. The third method is based on local structure prediction and it is able to segment the whole tumor region as well as tumor core and active tumor. This method takes the advantage of a fact that most medical images feature a high similarity in intensities of nearby pixels and a strong correlation of intensity profiles across different image modalities. One way of dealing with -- and even exploiting -- this correlation is the use of local image patches. In the same way, there is a high correlation between nearby labels in image annotation, a feature that has been used in the ``local structure prediction'' of local label patches. Convolutional neural network is chosen as a learning algorithm, as it is known to be suited for dealing with correlation between features. All three methods were evaluated on a public data set of 254 multisequence MR volumes being able to reach comparable results to state-of-the-art methods in much shorter computing time (order of seconds running on CPU) providing means, for example, to do online updates when aiming at an interactive segmentation.
|
|
|
Automatic detection of brain tumors in MR images
Dvořák, P. ; Kropatsch, W.G. ; Bartušek, Karel
This paper deals with automatic brain tumor detection in magnetic resonance images. The goal is to determine whether the MR image of a brain contains a tumor. The proposed method works with T2-weighted magnetic resonance images, where the head is vertically aligned. The detection is based on checking the left-right symmetry of the brain, which is the assumption for healthy brain. The algorithm was tested by fivefold cross-validation technique on 72 images of brain containing tumors and 131 images of healthy brain. The proposed method reaches the true positive rate of 91.16% and the true negative rate of 94.68%.
|
host ::
RSS.