 Deep learning algorithms have been used to analyze electronic health records, EHRs, in order to identify type 2 diabetes, T2D. This study combined radiological and EHR data to develop a deep learning model which accurately identified T2D with a rock AUC of 0.84. The model also flagged 1,381 cases, 14%, as being suspected of having T2D. Further validation at another institution showed a rock AUC of 0.77, indicating that the model could be used to screen for T2D. Additionally, the model's performance was explained through correlation analysis, revealing that certain adiposity measures were associated with higher predictive power. This article was authored by IE's pyros, Stephen M. Borstelman, Romana Mantravadi, and others.