Fig. 3
From: Prediction of longitudinal outcomes and novel cluster identification in epilepsy
Clusters from unsupervised learning. (A) Hierarchical Density-Based Spatial Clustering of Applications clustering with Noise (HDBSCAN) with t-distributed stochastic neighbor embedding (t-SNE), (B) five clusters were related to final outcome. Three clusters have better outcomes and two clusters have worse outcomes. These clusters exhibit distinct longitudinal outcomes. Cluster 1 demonstrates low-frequency seizures with delayed remission. Cluster 2 achieves early remission. Cluster 3 shows intermediate-frequency seizures with delayed remission. Cluster 4 experiences decreasing frequency but persistent seizures. Cluster 5 maintains a continuing high seizure frequency.
