|
Convolutional neural networks (CNNs) are a specialized class of deep neural networks. In the present era, these
have emerged as highly effective tools for a variety of computer vision tasks. Nonetheless, for classification tasks,
the application of a single CNN model is often not sufficient to achieve high precision and robustness. Ensemble
learning is a machine learning technique that can improve classification performance through combining multiple
models into one. With this method, individual models exchange each other’s best performance for each class,
resulting in improved overall accuracy. In this work, we studied the performance of CNN models for brain tumor
classification. As an outcome, we propose a novel ensemble CNN model for this purpose. We utilized the data-
set comes from Nanfang Hospital, which include 3064 MRI images categorized into three types of brain tumor
(glioma, meningioma and pituitary). First, we assessed well-known CNN models for their ability to classify brain
tumors. Next, we tested several ensemble transfer learning models and created one that utilizes the strengths of
the most efficient CNN models. The comparative analysis of model performance demonstrated that the examined
ensemble CNN models outperformed all single models. Moreover, regarding evaluation metrics, the proposed
model achieved global accuracy of 94% and the highest precision and recall, the F1 score of being 94%. Experi-
mental results revealed that model architecture and ensemble methods have a significant impact on brain tumor
classification performance.
|
