Hyperparameter Optimization of Convolutional Neural Networks for Robust Tumor Image Classification
Syed Muddusir Hussain1, Jawwad Sami Ur Rahman1, Faraz Akram1
1Biomedical Engineering Department, Riphah International University, I-14 Campus, Islamabad 45210, Pakistan.
Diagnostics (Basel, Switzerland)
|May 4, 2026
Summary
Optimized deep learning models improve brain tumor classification from MRI scans. This CNN achieved 95.35% accuracy with lower computational cost, outperforming deeper networks.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Brain tumors pose significant diagnostic challenges.
- Accurate tumor identification via Magnetic Resonance Imaging (MRI) is crucial for treatment.
- Developing efficient classification models is a key medical requirement.
Purpose of the Study:
- To design and optimize a Convolutional Neural Network (CNN) model for brain tumor classification using MRI.
- To enhance classification accuracy and reduce computational cost compared to existing architectures.
Main Methods:
- An optimized CNN model with dropout layers and hyperparameter tuning was developed.
- A dataset of 640 MRI scans (320 tumor, 320 non-tumor) was utilized.
- The model was trained using the Adam optimizer with a learning rate of 0.001 and benchmarked against VGG-19, Inception V3, ResNet-10, and ResNet-50.
Main Results:
- The optimized CNN achieved a peak training accuracy of 97.77% and a test accuracy of 95.35%.
- The model demonstrated superior performance and significantly lower training time compared to deeper architectures.
- High validation stability was observed, ranging from 92.25% to 95.35%.
Conclusions:
- Hyperparameter optimization and regularization are more critical than model depth for MRI tumor classification.
- The lightweight CNN offers a feasible solution for real-time applications due to its high accuracy and low computational complexity.