Brain Tumor Classification Using CNN and Channel Attention

Official materials and project page for our peer-reviewed article on multiclass brain tumor classification from MRI.

Article

Convolutional Neural Network and Channel Attention Mechanism for Multiclass Brain Tumor Classification Naderi A., Asgharzadeh-Bonab A., Ahmadi F., Kalbkhani H.

Journal	Complexity (Wiley), Volume 2025
Published	30 June 2025
Access	Open Access (CC BY 4.0)
DOI	10.1155/cplx/1644859

Abstract

This paper presents a deep learning framework combining Convolutional Neural Networks with a Channel Attention Mechanism for effective multiclass classification of brain tumors using MRI images. The proposed model comprises three components: a fine-tuned EfficientNetB7 backbone adapted through transfer learning, a channel attention module that refines extracted feature maps to emphasise clinically relevant tumour features, and a fully connected classifier optimised through grid search. Hyperparameter tuning and data augmentation further improve generalisation and robustness.

Results

Dataset	Task	Accuracy
Brats-4C	Four-class (glioma, meningioma, pituitary, no tumour)	98.16%
Brats-2C large	Binary	99.4%
Brats-2C small	Binary	99.2%

Validated with 5-fold stratified cross-validation.

Comparison with prior work

Method	Architecture	Accuracy
Kang et al. (2021)	DenseNet-169 + ShuffleNet + MnasNet	91.58%
Irmak (2021)	Custom CNN	92.66%
Shahin et al. (2023)	MPCANet (PCANet + CNN)	94.02%
Demir and Akbulut (2022)	R-CNN + SVM	96.60%
This work	EfficientNetB7 + CAM + FC	98.16%

Architecture

Transfer learning. EfficientNetB7 pre-trained on ImageNet. The first four MBConv blocks are frozen; blocks 5 to 7 are fine-tuned for domain-specific oncological patterns. Channel attention module. Exploits inter-channel relationships through average and max pooling to amplify clinically relevant tumour features before classification. Classifier head. Batch normalisation, a 256-neuron dense layer selected by grid search, 45% dropout, and softmax output. Training. SGDM optimiser, cross-entropy loss, L1 and L2 regularisation, and rigorous data augmentation.

Repository contents

File	Description
Project page	Live summary page
Complexity_2025_Naderi_Convolutional_Neural_Network_and_Channel.pdf	Full open-access article
Deep Learning & Medical Imaging.html	Extended case study with visualisations
index.html	Source of the project page

Citation

@article{naderi2025cnn,
  title   = {Convolutional Neural Network and Channel Attention Mechanism for Multiclass Brain Tumor Classification},
  author  = {Naderi, Ali and Asgharzadeh-Bonab, Akbar and Ahmadi, Farid and Kalbkhani, Hashem},
  journal = {Complexity},
  volume  = {2025},
  year    = {2025},
  publisher = {Wiley},
  doi     = {10.1155/cplx/1644859}
}

Authors

Ali Naderi - Department of Mechatronics Engineering, Urmia University of Technology ORCID: 0009-0004-8166-5449 | Portfolio | alinaderi119@gmail.com Akbar Asgharzadeh-Bonab - Department of Electrical and Computer Engineering, Urmia University Farid Ahmadi (corresponding author) - Department of IT and Computer Engineering, Urmia University of Technology Hashem Kalbkhani - Department of Electrical Engineering, Urmia University of Technology

License

The article is published Open Access under CC BY 4.0. Please cite the paper if you build on this work.