Grading Brain Histopathological Images Using Deep Residual Networks and Support Vector Machine

Yurttakal A. H., Erbay H., Arslan R. S.

Electronic Letters on Science and Engineering, vol.16, no.2, pp.77-83, 2020 (Peer-Reviewed Journal)

  • Publication Type: Article / Article
  • Volume: 16 Issue: 2
  • Publication Date: 2020
  • Journal Name: Electronic Letters on Science and Engineering
  • Journal Indexes: Index Copernicus
  • Page Numbers: pp.77-83
  • Kayseri University Affiliated: Yes

Abstract

: Brain cancer is a deadly disease that occurs due to tumour formation in the brain. It can cause weakness in the arms and legs, speech and vision disorders, extremely severe headaches, and symptoms such as vomiting. It is generally classified in four grades. The first and second grades are considered as "low grade", that is, "benign", and the third and fourth grade are considered as "high grade", that is, "malignant". The early grading of the tumour is important for the treatment procedures. Grading brain tumours based on histopathological images is a tiring process that requires expertise. On the other hand, deep learning algorithms are frequently used in computer-aided diagnostic systems. In this study, automatic grading of 1133x40 brain histopathologic images belonging to four phases was performed. First of all, features have been extracted from the latest technology pre-trained Residual networks with the ResNet50 and ResNet101 models. Then, the hyper parameters were optimized by Bayesian Optimization and classified by the Support Vector Machine (SVM) algorithm. 80% of the data set is reserved for training and 20% for testing. When evaluated in terms of multiple classification problems, it is reached a high accuracy rate of 80.09% in ResNet50, while reaching 100% high recall value in Grade I detection in Resnet101.