An Approach of Deep Learning Cascaded Model for Brain Tumor Segmentation on MRI Images

Authors

  • Archana K V, Komarasamy G

Abstract

Computer Aided Diagnosis (CAD) systems are widely adopted in the field of bio-medical where different types of images are captured and processed to improve the reliability and efficiency of diagnosis. MRI is considered as a promising technique because it provides the complete information about the human organ, thus it is used in various diagnosis systems. Currently, brain tumor is one of the leading causes for mortality in humans hence, early stage detection and prediction of brain tumor can help to reduce the mortality rate. Various techniques are presented during last decade that are mainly focused on the brain tumor segmentation and classification. However, the classification accuracy depends on the segmentation, hence to focus on brain tumor segmentation in this work. Nowadays, deep learning-based schemes have gained attraction for semantic segmentation. However, achieving the desired performance of accurate segmentation is still considered as a challenging task. In order to overcome the issue of tumor segmentation, introduce a deep learning-based scheme which is mainly based on the encoding and decoding process using cascading method. Introduce a new architecture of encoder and decoder module which uses 3x3 kernels with stride 2. Moreover, the proposed model also uses ResNet model for learning as a residual learning process. Proposed approach achieves average dice score as 0.95 which shows a significant improvement in contrast to existing techniques.

Keywords- Medical image processing, brain tumor segmentation, deep learning, computer vision

Published

2020-12-08

Issue

Section

Articles