From subjective assessment to data-driven diagnosis: the AI revolution in multimodal early detection of Sjögren's disease.

Sjögren's disease (SjD), a chronic autoimmune disorder affecting exocrine glands, faces significant diagnostic challenges due to its highly heterogeneous symptoms, subjective interpretation of imaging findings, and reliance on invasive biopsies, often resulting in delayed diagnosis by 3-7 years. This review posits that the integration of diverse data streams-from medical imaging to molecular omics-is the pivotal key to overcoming diagnostic heterogeneity, and that AI serves as the indispensable engine to power this integration. We systematically synthesize evidence showing that deep learning models, such as Fully Convolutional Dense Network (FCN-DenseNet), can automate salivary gland ultrasound segmentation; computed tomography (CT)-based AlexNet may achieve diagnostic performance comparable to radiologists; and AI-driven analyses of pathology, metabolomics, and genomics have revealed potential biomarker patterns, including metabolomic and immune-clustering models with reported AUC values of 1.00. However, these findings should be interpreted cautiously because many studies were conducted in small, retrospective, single-center cohorts or relied mainly on internal validation, limiting their clinical robustness and generalizability. We thus critically advocate for a future paradigm centered on causally-aware, multimodal fusion frameworks-moving beyond mere correlation to model disease mechanisms-and the adoption of federated learning to navigate data privacy while leveraging large-scale genomics. The forthcoming challenge is not merely technical but conceptual: to develop interpretable AI that can deconstruct SjD into mechanistically defined subtypes, thereby bridging the critical gap between algorithmic prowess and clinically actionable insights for definitive early intervention and personalized treatment strategies.