Differentiating Tumor and Edema in Brain Magnetic Resonance Images using a Convolutional Neural Network

  • Aida Allahverdi Dipartimento di Seienze dell'Informazione Universita" 'La Sapienza' di Roma, Via Salaria 113, 00198 Roma, Italy
  • Siavash Akbarzadeh Dipartimento di Seienze dell'Informazione Universita" 'La Sapienza' di Roma, Via Salaria 113, 00198 Roma, Italy
  • Alireza Khorrami Moghaddam Radiology Department, Sari School of Allied Medical Sciences, Mazandaran University of Medical Sciences
  • Armin Allahverdy Radiology Department, Sari School of Allied Medical Sciences, Mazandaran University of Medical Sciences
Keywords: Magnetic Resonance Image, Segmentation, Convolutional Neural Network, Glioblastoma

Abstract

Purpose- Glioblastoma is the most common subset of glioma with high grade of mortality. Therefore, early diagnosis may cause the better therapeutic interventions. Moreover, brain MRI shows good performance to tumor localization. But manual tumor localization is time consuming therefore an automatic tumor segmentation is recommended. Method- In this study, an automatic brain tumor segmentation based on the fully convolutional neural network is presented. This segmentation method can localize and differentiate the active tumor and edema in multi-modal brain MRI. The convolutional neural network has a wide range application for machine vision and visual recognition. In this study, we introduced a novel convolutional neural network for brain tumor segmentation. Results- This method was used for high grade and low grade subjects’ MRI. In this study a multi-modal MRI data contained T1 weighted, T1 enhanced, T2 weighted and FLAIR was used. For investigating the segmentation performance, the dateset was divided into train and test dataset. Moreover the fully convolutional neural network used the pixels of sliding window on MRI as input. The results shows that the increasing the window size cause the increment of train segmentation performance and has no significant effect on segmentation performance. Overall, the best train segmentation performance was 97.6% and best test segmentation performance was 89.7%.
Published
2018-09-01
How to Cite
1.
Allahverdi A, Akbarzadeh S, Khorrami Moghaddam A, Allahverdy A. Differentiating Tumor and Edema in Brain Magnetic Resonance Images using a Convolutional Neural Network. FBT. 5(1-2):44-0.
Section
Original Article(s)