<p>Capillaroscopy remains one of the most critical non-invasive diagnostic tools for the assessment of peripheral microvascular abnormalities in rheumatology and autoimmune disorders. Recent advancements in deep learning, particularly object detection models such as YOLO (You Only Look Once), offer transformative potential in automating capillaroscopy image interpretation. This study presents a comprehensive comparative evaluation of five state-of-the-art YOLO variants—YOLOv8, YOLOv9, YOLOv10, YOLOv11, and YOLOv12—for the classification and localization of capillaroscopic patterns. Utilizing a curated dataset of 321 annotated nailfold capillaroscopy images, the models were evaluated based on diagnostic accuracy (mAP@0.5, mAP@0.5–0.95, precision, recall) and inference time per image. Results indicate that YOLOv9 achieves the highest accuracy across all mAP and precision metrics, establishing itself as the most diagnostically reliable model. YOLOv11 offers the highest recall, favoring sensitivity and minimizing false negatives, which are crucial in early autoimmune disease screening. YOLOv8 is distinguished by its real-time performance, processing each image in approximately 4&#xa0;ms, ideal for portable and embedded healthcare systems, while YOLOv10 underperformed across all categories. The study includes expanded details on dataset provenance, annotation quality, augmentation parameters, and statistical validation. These refinements enhance reproducibility and transparency and clarify the clinical implications of each YOLO model’s behavior. This research bridges the gap between machine learning performance and clinical applicability in microvascular diagnostics and supports the adoption of deep-learning-powered capillaroscopy as a reliable and equitable diagnostic aid for early disease detection.</p>