Manuscript received January 5, 2026; revised January 30, 2026; accepted March 24, 2026; published July 21, 2026.

Abstract—Accurate brain tumor segmentation supports clinical decision-making, including treatment planning and surgical navigation, yet manual delineation is time-consuming and labor-intensive. Many high-accuracy models remain difficult to deploy under latency and memory constraints, while multimodal Magnetic Resonance Imaging (MRI) introduces severe foreground-background imbalance. Cloud-Net is a lightweight U-shaped convolutional framework that places a Spectral-Spatial Feature Enhancer (SFFE) at the bottleneck, combining a spatial convolution branch with Fourier-domain feature modulation to strengthen global structural cues while preserving local boundary detail. The model is optimized with a hybrid Dice-Focal objective and evaluated on the Brain Tumor Segmentation 2020 (BraTS 2020) dataset using T1-weighted (T1), contrast-enhanced T1-weighted (T1ce), T2-weighted (T2), and Fluid-Attenuated Inversion Recovery (FLAIR) MRI sequences. On tumor present slices, Cloud-Net achieves a Dice Similarity Coefficient (DSC) of 0.7701 and an Intersection over Union (IoU) of 0.6825, with a 95th-percentile Hausdorff Distance (HD95) of 20.21 pixels, while maintaining real-time efficiency at 2.73 ms/slice and 365.64 Frames Per Second (FPS) using 9.35 million parameters (35.69 MB). Ablation analysis indicates that SFFE improves overlap and boundary quality, and Gradient-weighted Class Activation Mapping (Grad-CAM) highlights lesion-relevant regions influencing predictions. These results support Cloud-Net as a practical option for real-time brain tumor segmentation in cloud-edge-oriented workflows.

Keywords—imaging, lightweight convolutional neural network, spectral-spatial fusion, gradient-weighted class activation mapping
 

Cite: Jyoti Kataria and Supriya P. Panda, "Cloud-Net: A Lightweight Convolutional Framework with Spectral-spatial Feature Enhancer for Real-time Brain Tumor Segmentation," Journal of Image and Graphics, Vol. 14, No. 4, pp. 687-705, 2026.

Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).