An Explainable model for diagnosing different grades of Brain Tumor | ||||
| Engineering Research Journal (Shoubra) | ||||
| Volume 54, Issue 2, April 2025, Page 7-19 PDF (1.75 MB) | ||||
| Document Type: Research articles | ||||
| DOI: 10.21608/erjsh.2025.372558.1404 | ||||
 View on SCiNiTO | ||||
| Authors | ||||
Hala Badr Kamal  ; Lamiaa Elrefaei ; Eman Abdel-Ghaffar
| ||||
| Electrical Engineering Department, Faculty of Engineering at Shoubra, Benha University, Cairo, Egypt | ||||
| Abstract | ||||
| A brain tumor is characterized by the uncontrolled growth of cells in the brain due to genetic mutations in the deoxyribonucleic acid (DNA) of these cells’ genes. There are four grades of brain tumor development. So the doctor must detect the grade to which the patient has arrived to develop the appropriate treatment plan but this tumor grade detection process is difficult. For this reason, in this research we propose an accurate detection of brain tumor disease utilizing various deep learning and machine learning techniques. In this research two publicly available datasets used: FigShare and Rembrandt dataset. The pre-trained MobileNet model was trained on only axial magnetic resonance imaging (MRI) images from the two datasets. The pre-trained MobileNet was fine-tuned for feature extraction. Then applied ML models (Random Forest, Support vector machine and Decision tree) to classify brain tumor types from the first dataset and brain tumor grades from the second dataset. Which achieved superior accuracy compared to other models 99% and 99.74% accuracy for the two datasets. To explain the results from the pre-trained MobileNet model, two types of XAI methods were applied: Grad-CAM and Shap. | ||||
| Keywords | ||||
| Brain Tumor (BT); classification; explainable artificial intelligence (XAI); Gradient-weighted class activation mapping (Grad-CAM); SHapley Additive exPlanations (SHAP) | ||||
|
Statistics Article View: 141 PDF Download: 58 |
||||
