Figure 1.

Design of the analysis. (i) To provide baseline results, a clinical Cox proportional hazards model (CPHM) was trained on the exploratory cohort and evaluated on the independent validation cohort. (ii)–(iv) Three deep learning approaches were evaluated by training convolutional neural networks in a cross-validation approach. Subsequently, for each approach ensembles were constructed from the models obtained during cross-validation and their performance was evaluated on the independent validation cohort.