CBAM-ConvNeXt: Enhancing ConvNeXt With Channel and Spatial Attention for Accurate Brain Tumour Classification

Abstract

Accurate diagnosis of brain tumours remains a critical challenge due to their morphological heterogeneity and overlapping visual characteristics. This study proposes CBAM-ConvNeXt, a deep convolutional framework that integrates the Convolutional Block Attention Module (CBAM) into the ConvNeXt backbone to enhance both channel-wise and spatial feature representation. The model adaptively highlights diagnostically salient regions in MRI scans while suppressing redundant information. Experiments conducted on a publicly available brain tumour MRI dataset with 7,023 contrast-enhanced T1-weighted images across four tumour classes demonstrated that the proposed method achieved 0.9977 +/- 0.0006 accuracy, 0.9975 +/- 0.0006 F1-score, 0.9975 +/- 0.0006 precision, 0.9975 +/- 0.0006 recall and AUC = 1.0000 +/- 0.0000, averaged over five independent random seeds. Grad-CAM visualisations confirmed that the model focuses on clinically meaningful tumour regions. These findings suggest that integrating channel and spatialattention within ConvNeXt maintains high predictive accuracy while substantially improving interpretability in clinical MRI applications.