CapsNet topology to classify tumours from brain images and comparative evaluation. Issue 5 (4th March 2020)

• Record Type: Journal Article
• Authors: Goceri, Evgin
• Is Part Of: IET image processing. Volume 14:Issue 5(2020)
• Page Start: 882
• Page End: 889
• Publication Date: 2020-03-04
• Subjects: medical image processing -- image classification -- neural nets -- feature extraction -- brain -- biomedical MRI -- gradient methods -- learning (artificial intelligence) -- expectation‐maximisation algorithm -- tumours -- optimisation; CapsNet topology -- brain images -- brain tissues -- meningioma -- ependymoma -- computer‐assisted brain tumour classification techniques -- Capsule‐based neural networks -- tumour recognition -- brain magnetic resonance images -- glioma -- CapsNet based methods -- expectation‐maximisation based dynamic routing -- tumour boundary information -- pituitary -- Sobolev gradient‐based optimisation -- network topology -- feature extraction -- learned features; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/iet-ipr.2019.0312
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract : Visual evaluation of many magnetic resonance images is a difficult task. Therefore, computer‐assisted brain tumor classification techniques have been proposed.These techniques have several drawbacks or limitations. Capsule based neuralnetworks are new approaches that can preserve spatial relationships of learnedfeatures using dynamic routing algorithm. By this way, not only performance oftumor recognition increases but also sampling efficiency and generalisationcapability improves. Therefore, in this work, a Capsule Network (CapsNet) isused to achieve fully automated classification of tumors from brain magneticresonance images. In this work, prevalent three types of tumors (pituitary, glioma and meningioma) have been handled. The main contributions in this paperare as follows: 1) A comprehensive review on CapsNet based methods is presented.2) A new CapsNet topology is designed by using a Sobolev gradient‐basedoptimisation, expectation‐maximisation based dynamic routing and tumor boundaryinformation. 3) The network topology is applied to categorise three types ofbrain tumors. 4) Comparative evaluations of the results obtained by othermethods are performed. According to the experimental results, the proposedCapsNet based technique can achieve extraction of desired features from imagedata sets and provides tumor classification automatically with 92.65%accuracy.