PatTree Automates Multimodal Patient Data for Medical AI.

Artificial Intelligence in medicine often benefits from holistic, multimodal patient data, but the inherent complexity and heterogeneity of real-world clinical data—including missing values, varied time points, and inconsistent formats—pose significant challenges for structured analysis. Traditional data harmonization is resource-intensive and limits scalability. This research introduces PatTree, a novel graph-based representation that automates the structuring of multimodal clinical data to create comprehensive patient profiles. PatTree enables early-stage data integration without requiring pre-standardized inputs. By representing diverse clinical data within a unified knowledge graph, it preserves semantic relationships across modalities and sources, enhancing interoperability and machine-interpretability. The effectiveness of PatTree was demonstrated using a subset of the ADNI-1 cohort, where it achieved state-of-the-art classification performance. For a three-class classification task distinguishing Alzheimer's disease, mild cognitive impairment, and cognitively normal individuals, PatTree reached a balanced accuracy of 98.5% and an F1 score of 0.987 on a held-out test set, proving its potential to streamline clinical AI pipelines by eliminating tedious data preparation.