Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Deep Learning Approach for Brain Tumor Classification and Segmentation Using a Multiscale Convolutional Neural Network

Francisco Javier Díaz-Pernas, Mario Martínez-Zarzuela, Míriam Antón-Rodríguez, David González-Ortega

TL;DR

A fully automatic brain tumor segmentation and classification model using a Deep Convolutional Neural Network that includes a multiscale approach that is inspired in the inherent operation of the Human Visual System.

Abstract

In this paper, we present a fully automatic brain tumor segmentation and classification model using a Deep Convolutional Neural Network that includes a multiscale approach. One of the differences of our proposal with respect to previous works is that input images are processed in three spatial scales along different processing pathways. This mechanism is inspired in the inherent operation of the Human Visual System. The proposed neural model can analyze MRI images containing three types of tumors: meningioma, glioma, and pituitary tumor, over sagittal, coronal, and axial views and does not need preprocessing of input images to remove skull or vertebral column parts in advance. The performance of our method on a publicly available MRI image dataset of 3064 slices from 233 patients is compared with previously classical machine learning and deep learning published methods. In the comparison, our method remarkably obtained a tumor classification accuracy of 0.973, higher than the other approaches using the same database.

A Deep Learning Approach for Brain Tumor Classification and Segmentation Using a Multiscale Convolutional Neural Network

TL;DR

A fully automatic brain tumor segmentation and classification model using a Deep Convolutional Neural Network that includes a multiscale approach that is inspired in the inherent operation of the Human Visual System.

Abstract

In this paper, we present a fully automatic brain tumor segmentation and classification model using a Deep Convolutional Neural Network that includes a multiscale approach. One of the differences of our proposal with respect to previous works is that input images are processed in three spatial scales along different processing pathways. This mechanism is inspired in the inherent operation of the Human Visual System. The proposed neural model can analyze MRI images containing three types of tumors: meningioma, glioma, and pituitary tumor, over sagittal, coronal, and axial views and does not need preprocessing of input images to remove skull or vertebral column parts in advance. The performance of our method on a publicly available MRI image dataset of 3064 slices from 233 patients is compared with previously classical machine learning and deep learning published methods. In the comparison, our method remarkably obtained a tumor classification accuracy of 0.973, higher than the other approaches using the same database.
Paper Structure (7 sections, 2 equations, 7 figures, 4 tables)

This paper contains 7 sections, 2 equations, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Materials and Methods
  3. Proposed Convolutional Neural Network and Implementation Details
  4. Dataset and Data Preprocessing
  5. Neural Network Training and Performance Measurements
  6. Results and Discussion
  7. Conclusions

Figures (7)

  • Figure 1: Examples of MRI images of the T1-CE MRI image dataset. Left: coronal view of a meningioma tumor. Center: Axial view of a glioma tumor. Right: sagittal view of a pituitary tumor. Tumor borders have been highlighted in red.
  • Figure 2: The proposed Convolutional Neural Networks (CNN) architecture. Input:$1 \times 65 \times 65$ sliding windows. Model: Three pathways (large, medium, and small feature scales) with 2 convolutional layers and max-pooling, a convolutional layer with concatenation of the three pathways, and a fully connected stage that leads to a classification in one out of the four possible output labels: 0-healthy region, 1-meningioma tumor, 2-glioma tumor, and 3-pituitary tumor. A dropout mechanism between the concatenation and fully connected stages is included.
  • Figure 3: Example of elastic transformation used in the data augmentation. Left: original slice. Right: image transformed. In both images, the edges of the tumor have been highlighted in red.
  • Figure 4: Examples of results of the proposed method for three slices corresponding to meningioma, glioma, and pituitary tumors, respectively. The images show the tumor segmentation: The region detected is shown in red while the ground truth region is shown in green. As a result, the intersection region is shown in yellow.
  • Figure 5: Histograms of metrics obtained after processing the dataset. Left: Dice histogram. Center: Sensitivity histogram. Right: Histogram of the predicted tumor type.
  • ...and 2 more figures