Residual Attention Network for Brain Tumour Classification

• Main Author: Sashwini, A/P S. Thiagaraju
• Format: Final Year Project Report
• Language: English; English
• Published: Universiti Malaysia Sarawak (UNIMAS) 2019
• Subjects: H Social Sciences (General); RC0254 Neoplasms. Tumors. Oncology (including Cancer)
• Online Access: http://ir.unimas.my/id/eprint/27561/1/Residual%20attention%20network%20for%20brain%20tumor%20detection%20%2824%20pgs%29.pdf; http://ir.unimas.my/id/eprint/27561/4/Sashwini%20ft.pdf; http://ir.unimas.my/id/eprint/27561/

The main aim of this study is to design and produce an automated algorithm system using Residual Attention Network (RAN) model, which will classify brain tumour. In this project digitalised Magnetic Resonance Image (MRI) is used which is obtained from Malaysian hospitals. The MRI dataset consists of those of patients who are 20 years and above both male and female. The Residual Attention Network model is trained and tested using the MRI dataset. The performance of the algorithm is evaluated based on training accuracy, testing accuracy, validate accuracy and validate loss and comparative analysis with Residual Neural Network (ResNet) and Convolutional Neural Network (CNN). ResNet and CNN were tested using the same dataset. Results from this study certainly proved that RAN provided the best performance among the 3 algorithms. ResNet has good performance with its accuracy ranging from 66% to 90%. The normal CNN algorithm did not perform well with the accuracy being very inconsistent between 57% and 71 %. RAN produced the highest and most consistent accuracy which is from 94% onwards. Further explanation is provided to prove the efficiency of Residual Attention Network for the classification of brain tumour.