|
|
 |
| |
|
|
|
Hyber Selective Ensemble Methodology Based On Deep Transfer Learning For Brain Diagnosis Detection |
|
|
|
PP: 9-19 |
|
|
|
doi:10.18576/aeta/110201
|
|
|
|
Author(s) |
|
|
|
Amal fouad Abd El-Hady,
Hesham Ahmed Hefny,
Rowayda Abd El-hamid Sadek,
Hossam M. Moftah,
|
|
|
|
Abstract |
|
|
| Machine Learning models initiate to have a great effect on the diagnosis of numerous diseases. In the biomedical field, Convolutional neural networks (CNNs) display a potential role for computer-aided diagnosis (CAD) by extracting features directly from the image data instead of the features based on analytically methods or handcrafts features. However, CNNs have many challenges to train medical images from scratch as small sample sizes and variations in tumor presentations. Additionally, it needs more hardware for processing. Alternatively, transfer learning can extract from medical images tumor information through CNNs originally pre-trained for nonmedical images, which cover the shortage of a small dataset.
The proposed model introduces several pre-trained models such as Xception, VGG16, VGG19, ResNet50, MobileNet, MobileNetV2, and InceptionResNetV2 to create a selective ensemble model from them which achieves 97.77accuracy on brain tumor type classification.
|
|
|
|
|
|
|
|
