Local Structure Prediction with Convolutional Neural Networks for Multimodal Brain Tumor Segmentation

Local Structure Prediction with Convolutional Neural Networks for Multimodal Brain Tumor Segmentation

Článek WoS

Most medical images feature a high similarity in the 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. In the present study we test this local structure prediction approach for 3D segmentation tasks, systematically evaluating different parameters that are relevant for the dense annotation of anatomical structures. We choose convolutional neural network as learning algorithm, as it is known to be suited for dealing with correlation between features. We evaluate our approach on the public BRATS2014 data set with three multimodal segmentation tasks, being able to obtain state-of-the-art results for this brain tumor segmentation data set consisting of 254 multimodal volumes with computing time of only 13 seconds per volume.

Most medical images feature a high similarity in the 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. In the present study we test this local structure prediction approach for 3D segmentation tasks, systematically evaluating different parameters that are relevant for the dense annotation of anatomical structures. We choose convolutional neural network as learning algorithm, as it is known to be suited for dealing with correlation between features. We evaluate our approach on the public BRATS2014 data set with three multimodal segmentation tasks, being able to obtain state-of-the-art results for this brain tumor segmentation data set consisting of 254 multimodal volumes with computing time of only 13 seconds per volume.

Brain Tumor, Clustering, CNN, Deep Learning, Image Segmentation, MRI, Patch, Structure, Structured Prediction.

Brain Tumor, Clustering, CNN, Deep Learning, Image Segmentation, MRI, Patch, Structure, Structured Prediction.

DVOŘÁK, P.; MENZE, B.

2016

09.10.2015

Proceedings MICCAI-MCV 2015

0302-9743

Lecture Notes in Computer Science

8965

1

Spolková republika Německo

1

12

12