Skip to main content

Table 5 External validation results.

From: A deep neural network approach to predicting clinical outcomes of neuroblastoma patients

DatasetsBalanced accuracy
TrainingTestDNNSVMRF
Clinical outcome = ‘Death from disease’, 
Data = centralities 
Fischer-MFischer-M87.3%75.4%75.1%
 Fischer-R82.1%53.5%66.8%
 Maris53.1%54.3%50.0%
 Versteeg75.0%53.3%67.5%
Fischer-RFischer-R85.8%66.0%62.4%
 Fischer-M81.5%75.4%61.2%
 Maris56.2%49.7%50.0%
 Versteeg70.8%68.3%67.5%
Clinical outcome = ‘Disease progression’, 
Data = centralities 
Fischer-MFischer-M84.3%83.7%80.0%
 Fischer-R77.0%75.2%71.8%
 Maris67.5%66.0%53.8%
 Versteeg78.1%82.4%78.1%
Fischer-RFischer-R83.7%81.0%73.3%
 Fischer-M80.0%76.8%75.0%
 Maris67.5%58.8%58.8%
 Versteeg80.1%77.2%73.9%
  1. Models are trained using one of the ‘Fischer’ datasets and then tested using either the other ‘Fischer’ dataset or another independent dataset (‘Maris’ and ‘Versteeg’). The ‘Maris’ and ‘Versteeg’ datasets are too small to be used for both training and therefore are only used for validation. Rows in italics represent reference models (training and testing extracted from the same datasets)