Optimized Data Augmentation and Lightweight Models for Brain Tumor Segmentation and Classification in Low–Resource Settings

Accurate classification and segmentation of brain tumors are crucial in medical imaging, yet they present significant challenges, particularly in resource–limited settings with limited data and computational power. This study tackles these issues through two primarystrategies: adaptive data augmentation and lightweight neural network models. Adaptive data augmentation improves the diversity and quality of medical images using techniques such as geometric transformations, intensity adjustments, and synthetic data generation. By customizing these augmentations to specific tumor characteristics and imaging conditions, this approach enhances model performance even with smaller datasets. To facilitate real–time tumor classification and segmentation while minimizing computational demands, we developed lightweightneural network models. These models utilize streamlined architectures, including depthwise separable convolutions and attention mechanisms, achieving high accuracy and efficiency, making them well–suited for deployment in low–resource environments. Additionally, quantizationand pruning techniques effectively reduce the size and complexity of the neural networks without compromising performance. Experimental results indicate that these methods significantly improve segmentation accuracy and classification precision across various MRI datasets. This research underscores the potential of adaptive augmentation and efficient neural network designs to provide advanced diagnostic capabilities in resource–constrained settings. Futurework will investigate the integration of these techniques with federated learning and explainable AI to further enhance reliability and clinical applicability.