DeepLesionBrain performs fully automated lesion segmentation from T1-weighted and FLAIR images, enabling robust analysis from standard clinical MRI acquisitions. The method relies on an ensemble of compact 3D convolutional neural networks trained with advanced data augmentation strategies, allowing the system to generalize across heterogeneous MRI datasets, scanners, and acquisition protocols. A key innovation of the framework is hierarchical specialization learning, where a global model first learns general lesion-related patterns across the brain. This model then initializes a set of region-specific networks that focus on different anatomical areas. By combining global contextual information with localized lesion patterns, DeepLesionBrain achieves accurate and robust segmentation performance. The pipeline also integrates automatic preprocessing steps, including denoising, bias-field correction, and multimodal registration between T1 and FLAIR images. After segmentation, the system computes quantitative lesion metrics, such as total lesion volume and spatial distribution across anatomical regions (e.g., periventricular, deep white matter, or juxtacortical areas). By providing fast, reproducible, and fully automated lesion quantification, DeepLesionBrain facilitates large-scale neuroimaging studies and supports research in neurological conditions such as multiple sclerosis, small vessel disease, and brain aging.