Running summarizing tools on a new article
Check the paper below, then start its synthesis.
Interpretable Deep Learning for Automated MRI-Based Brain Tumor Diagnosis Using MobileNet and DenseNet
Ratchanon Pawanja (Algoma University), Janelle Carpena (Algoma University), Rachelle Chan (Algoma University), Ho Toan Pham (Algoma University), Khalil Al-Hussaeni (Rochester Institute of Technology), Rashid Hussain Khokhar (Algoma University)
Abstract: Brain tumor classification using Magnetic Resonance Imaging (MRI) is vital for accurate diagnosis and effective treatment planning. Manual analysis of MRI scans is often labor-intensive and may vary depending on the radiologist's expertise. This research explores the application of deep learning approaches, particularly Convolutional Neural Networks (CNNs), for classifying brain tumors into four categories: glioma, meningioma, pituitary tumor, and no tumor. Two state-of-the-art CNN architectures, MobileNet and DenseNet, were fine-tuned using optimization techniques such as batch normalization, dropout, and adaptive learning rate scheduling. The models were trained and validated on a publicly available dataset consisting of 3,064 images, with consistent preprocessing and data augmentation applied to improve generalization. Performance was evaluated using accuracy, F1-score, and confusion matrices. Experimental results show that the proposed methods achieved 98.71% accuracy for MobileNet and 95.71% for DenseNet. These results demonstrate that optimized CNN architectures can achieve high classification performance while maintaining clinical relevance and interpretability. This work presents a practical, interpretable, and high-performing deep learning solution for multi-class brain tumor classification using MRI images.