Segmentation of Glioma Tumors in Brain Using Deep Convolutional Neural Network (1708.00377v1)

Abstract: Detection of brain tumor using a segmentation based approach is critical in cases, where survival of a subject depends on an accurate and timely clinical diagnosis. Gliomas are the most commonly found tumors having irregular shape and ambiguous boundaries, making them one of the hardest tumors to detect. The automation of brain tumor segmentation remains a challenging problem mainly due to significant variations in its structure. An automated brain tumor segmentation algorithm using deep convolutional neural network (DCNN) is presented in this paper. A patch based approach along with an inception module is used for training the deep network by extracting two co-centric patches of different sizes from the input images. Recent developments in deep neural networks such as drop-out, batch normalization, non-linear activation and inception module are used to build a new ILinear nexus architecture. The module overcomes the over-fitting problem arising due to scarcity of data using drop-out regularizer. Images are normalized and bias field corrected in the pre-processing step and then extracted patches are passed through a DCNN, which assigns an output label to the central pixel of each patch. Morphological operators are used for post-processing to remove small false positives around the edges. A two-phase weighted training method is introduced and evaluated using BRATS 2013 and BRATS 2015 datasets, where it improves the performance parameters of state-of-the-art techniques under similar settings.

• The paper introduces a novel patch-based deep CNN that improves segmentation accuracy of glioma tumors from multi-modal MRI images.
• It employs an innovative ILinear nexus architecture featuring inception modules, dropout, and batch normalization to optimize feature extraction.
• Results on BRATS 2013 and 2015 datasets show enhanced dice scores and computational efficiency, paving the way for improved clinical diagnosis.

Evaluation of Glioma Tumor Segmentation Using Deep Convolutional Neural Networks

The paper "Segmentation of Glioma Tumors in Brain Using Deep Convolutional Neural Network" addresses the critical task of automating the segmentation of glioma brain tumors from MRI data using deep learning technologies. Gliomas are characterized by their irregular shapes and indistinct boundaries, presenting a significant challenge for automated detection and segmentation processes. This paper contributes to the field by developing a deep convolutional neural network (DCNN) approach, vital for enhancing the accuracy and timeliness of clinical diagnoses in brain tumor cases.

Methodology Overview

The research emphasizes an innovative patch-based DCNN architecture that leverages an "ILinear nexus" framework for glioma tumor segmentation. The network integrates modern deep learning techniques such as inception modules, drop-out regularization, batch normalization, and non-linear activation functions. It systematically processes multiple co-centric patches extracted from multi-modal MRI images (T1, T1c, T2, T2-Flair) to segment tumor regions. The authors provide a comprehensive three-step methodology comprising pre-processing, convolutional network processing, and post-processing phases to enhance image segmentation accuracy.

Data and Experimental Setup

The proposed DCNN architecture was evaluated utilizing the publicly available BRATS 2013 and BRATS 2015 datasets. These datasets provide a benchmark for testing brain tumor segmentation techniques with standardized evaluation metrics, facilitating direct comparisons to previous state-of-the-art models. Pre-processing involves bias field corrections and intensity normalization to ensure consistent inputs to the network.

Results and Analysis

The paper reports that the new ILinear nexus architecture achieves significant improvements over existing methods, particularly in the segmentation accuracy of core and enhancing tumor regions, exhibiting high dice scores on both datasets. Utilizing a two-phase weighted training process was instrumental in addressing the class imbalance problem prevalent in medical imaging, where tumor pixels are considerably fewer compared to healthy tissue pixels. Overall, the results signify that the architecture not only enhances segmentation precision but also improves computational efficiency, reducing the time required for processing each brain from state-of-the-art methods.

Implications and Future Directions

The implications of this research are extensive, offering a potent tool for medical diagnostics, potentially leading to improved treatment planning and clinical outcomes for glioma patients. It also opens avenues for further research in optimizing neural network architectures for medical imaging applications, such as integrating 3D volumetric data for a more comprehensive analysis. Furthermore, the findings underscore the importance of addressing data imbalance and feature generalization in medical datasets, suggesting a potential shift towards more robust, generalizable DCNN models in the field of medical image analysis.

In conclusion, this paper offers a substantive advancement in brain tumor segmentation using deep learning, presenting methodologies and results that could significantly influence future research and practical applications in clinical settings.