Self-Organizing Maps
Online data visualization using the neural gas network
Neural Networks - 2006 Special issue: Advances in self-organizing maps--WSOM'05
The parameterless self-organizing map algorithm
IEEE Transactions on Neural Networks
Hi-index | 0.00 |
Unsupervised K-means cluster analysis and self-organizing maps (SOM) were employed to cluster patients based on feature values in the large Digitalis Investigation Group (DIG) trial database of digoxin for heart failure treatment. We observed that use of standardized features for input into SOM resulted in clusters for which the pattern of features were much different from clusters obtained using K-means and SOM with normalized features. Cox proportional hazards regression modeling allowed us to identify clusters whose subjects had increased all-cause mortality risk due to digoxin treatment. Results indicate that increased all-cause mortality risk with digoxin treatment was associated with female gender, older age, systolic blood pressure, heart rate, body mass index, CT ratio, ejection fraction, history of diabetes mellitus, history of hypertension, diuretic use, and less prevalence of a third heart sound. Combined use of cluster analysis and Cox regression identified an association with increased risk of all-cause mortality with treatment of digoxin in certain heart failure patients.