Neural Networks for Survival Prediction in Medicine Using Prognostic Factors: A Review and Critical Appraisal

Figure 2

Two basic architectures of survival neural networks. (a) A network where time (interval) is coded as a prognostic variable (input feature). Data transformation into a long format is required for each patient. The output layer makes predictions in a given time interval. (b) A network where time (interval) is not coded as part of the prognostic variables. The wide data format is adequate for each patient. The output layer makes predictions at multiple sequential (nonoverlapping) time intervals.

(a)

(b)