Developing a diagnostic system for brain tumor types in MRIs using self-organizing maps and feed forward artificial neural networks

Abstract

Abnormal cell growth leads to a tumor in the brain cells. Early detection, diagnosis and treatment of brain tumor is necessary to prevent human death. In this research, a diagnostic system was proposed to detect and classify the type of brain tumor among three types in the magnetic resonance images. The proposed algorithm first relied on the use of self-organizing maps (SOM) to segment the image and detect the tumor area, followed by image processing and improvement processes and extract tumor properties, and finally, a Feed Forward neuronal network was used to diagnose the type of tumor based on the detected properties. A database containing 3064 improved magnetic resonance images has been adopted, each image contains tumor information, type, and location. The algorithm was trained on 300 images distributed equally in three types of brain tumor and 100 images without tumor. The results showed the accuracy and reliability of the proposed system, as 297 out of 300 images were successfully detected and diagnosed with sensitivity, specificity and accuracy reached 99%, 98% and 98% respectively.