Automatic brain tumour detection using image processing and data mining techniques

In recent days, extensive analysis on magnetic resonance imaging (MRI) is being performed to understand the complex structure of human brain. Generally, pathological regions in the brain are identified through various MRI acquisition techniques. Depending upon the MRI technique specific regions may be exhibited distinctly than the other brain regions. These images are analysed computationally to identify the abnormal regions. In this work, high grade glioma images are utilised to detect the tumour regions in the brain using image processing and data mining techniques. Broadly, the pixels are grouped into tumour and non-tumour pixels using unsupervised as well as supervised data mining methods. Further, the tumour pixels are classified into four classes namely, oedema, necrosis, enhancing tumour and non-enhancing tumour using supervised classification methods. K-means clustering could detect the pixel clusters with an accuracy of 94.64% whereas random forest classifier could identify the pixel classes 99.50% correctly. Random forest could achieve better results in multi-label classification of the tumour when compared to other classifiers.