A Light Weight Deep Learning Framework for Brain Tumour Classification from Compressed MRI Images

The higher dimensionality of Magnetic Resonance Imaging (MRI) poses challenges for storage, transmission, and computational efficiency when used with deep learning (DL) models for automated tumour classification. This paper introduces a novel lightweight DL framework that uses MobileNetV2 to classify brain tumours from discrete wavelet transform (DWT) compressed brain MRI images. The proposed framework is developed and deployed on a Raspberry Pi 4B single-board computer and achieves an accuracy of $\mathbf{9 4. 0 8 \%}$ for compressed images, compared to $\mathbf{9 3. 7 1 \%}$ for uncompressed images. The performance analysis of the developed model was equally impressive, with compressed images outperforming uncompressed images in terms of precision (93.41% vs. $\mathbf{9 3. 0 9 \%}$), recall ($\mathbf{9 2. 9 7 \%}$ vs. $\mathbf{9 2. 8 1 \%}$), and F1-score (92.98% vs. $\mathbf{9 2. 9 1 \%)}$. The results demonstrate that DWT compressed MRI images consistently perform well with the model, achieving near-perfect classification for the ‘No Tumor’ class and robust performance for glioma, pituitary, and meningioma classes.