<p>Capillaroscopic image classification is a crucial tool for diagnosing microvascular diseases, such as systemic sclerosis. However, the task is challenging due to subtle capillary morphological variations, noise, overlapping characteristics, and the complexity of distinguishing healthy patterns from pathological ones. These challenges are amplified by the need to not only differentiate between healthy and unhealthy images but also to subclassify unhealthy images into disease-specific conditions, such as dystrophic patterns and stages of scleroderma (active, late, and early stages). To address these challenges, this research proposes a Capillaroscopic Multi-Level System Classification named HCap-ML-SC. The first level focuses on classifying images into healthy and unhealthy categories using a hybrid AI model that integrates ViT and ResNet architectures. This hybrid approach captures both local capillary details and global structural features, achieving a classification accuracy of 90.22%. The second level specializes in the subclassification of unhealthy images into detailed pathological subtypes, such as non-specific dystrophies and various stages of scleroderma, leveraging semantic learning methods. By incorporating expert annotations, the system achieved near to 100% accuracy in subclassifying these pathological subtypes. However, further validation is required to confirm generalizability. This dual-level system demonstrates the potential of hybrid models and semantic learning to enhance the accuracy and interpretability of capillaroscopic image classification. Future work will prioritize Explainable AI techniques to improve trust and transparency, further advancing clinical diagnostic tools.</p>