Authors

Maryam Bushra
Maymounah Abdullatif
Noura Alsady

Institutions

King Faisal University

Logo
  • Currently 3.00/5
  • 1
  • 1
  • 2
  • 2
  • 3
  • 3
  • 4
  • 4
  • 5
  • 5

Rating: 3.0/5 (42 votes cast)

Theme

Biomedical Engineering

Title

Computer-Aided Diagnosis of Brain Tumors from MR Images

Background

Summary of Work

Because of its importance for people in the healthcare community, we have tried to present an extensive review of the CAD systems as applied to automatic detection of brain tumor. This will help plan for future work in research and development that will improve the quality and efficiency of CAD systems. In the light of this review, we will represent the general CAD methodology for brain tumor detection using MRI images and its component steps (see figure below). The generic methodology of CAD system integrates different image processing techniques, which are pre-processing, image segmentation and feature extraction, and data processing techniques such as dimensionality reduction and classification.

 

  • Image acquisition and data collection ⇒ MRI images, mainly collected from the web repositories (due to legal and privacy issues).

 

 

  • Pre-processing  ⇒ makes the image data usable by enhancing it in terms of reduced noise and higher contrast and resolution. Median filter is one of the most applied methods of denoising.
  • Segmentation of the region of interest (ROI) ⇒ segregate an image image into mutually exclusive and exhausted regions which are homogeneous with respect to a predefined criterion. In specific case of brain tumors, it refers to separating tumor tissues from normal brain tissues.

 

Segmentation techniques based on the degree of human interaction
 Manual segmentation
Semi-automatic segmentation
Fully automatic segmentation

 

 

Algorithmic classification of segmentation techniques
Unsupervised segmentation
Supervised segmentation

 

  • Feature extraction ⇒ transform an image into a set of features for classification. 
  • Dimentionality reduction ⇒ reducing the extracted features into an optimum feature set with effective and discriminating features. 
  • Classification ⇒ classifying the input patterns (features) into analogous classes.
  • Performance evaluation ⇒ quantitative evaluation of the system and its performance using metrics of sensitivity, specificity, accuracy, and receiver operating characteristic (ROC) curve. 
  • The integration between feature extraction and feature reduction algorithms and the use of hybrid intelligent systems for classifier design led to developing CAD schemes with clinically acceptable accuracy and high efficiency, offering higher detection success rates. 

Conclusion

CAD systems and technology has contributed majorly in relatively quick and accurate diagnosis of brain tumor. Although many CAD systems that have been developed till date are robust and quick, there is still room for improvement that can enhance them in terms of accuracy and generalization. Most of the existing CAD systems suffer from drawbacks of high dimensionality of extracted features, high computational complexity and generalization capability. Therefore, CAD systems remain an open challenge mainly because they do not work in all cases due to different image quality and the size of the database. We suggest some considerations for future work which can further improve CAD systems based on MRI images such as acquisition of large databases from different resources with various image qualities and features, increased application of hybrid intelligent systems with soft computing techniques so as to resemble human reasoning, and improvement in classification accuracy by efficient feature extraction. For these reasons, there is a crucial need for extensive review of the literature in this area and it is strongly recommended to do state-of-the-art survey on the existing CAD systems to figure out their limitations so that new approaches are developed that overcome the drawbacks of existing research work in this area.

Background
Summary of Work
Conclusion
Send ePoster Link