Subject Area: BIOMETRIC TECHNOLOGY
This research presents a neural network-based approach for classifying brain tumors using magnetic resonance imaging (MRI). The primary objective is to develop a model capable of accurately detecting different classes of primary brain tumors. Review of relevant literature revealed that no existing solution captures all the primary brain tumor classes comprehensively. To address this gap, data collection was conducted, specifically targeting the primary brain tumor classes, including Medulloblastomas, gliomas, meningiomas, and pituitary adenomas tumors. Additionally, healthy brain tumor data was obtained from the Kaggle repository and integrated with the other classes for feature extraction. The extraction process employed a static and dynamic approach to convert the data into a compact feature vector. The feed-forward neural network algorithm was then trained using the extracted feature vector to generate the brain tumor classification model. The model's performance was evaluated using metrics such as accuracy, precision, and cross entropy through tenfold validation. The average error rate across the iterations was found to be 0.009814, indicating a low rate of misclassification. The average precision and accuracy were both determined to be 96.32%. Furthermore, a comparative analysis of the developed model against other classification approaches revealed an improvement of 3.76%. This suggests that the proposed neural network-based approach is more effective in accurately classifying primary brain tumors compared to existing methods. However, the limitation of the model is the issues of unbalance dataset which may result to classification bias in some cases. However recommendation using data augmentation or Adaboast algorithm can be used to address the probe in future research.