BME Frontiers / 2022 / Article / Fig 11

Research Article

Automated Segmentation and Connectivity Analysis for Normal Pressure Hydrocephalus

Figure 11

Test AUC for the linear SVM trained on different features: ventricle, subarachnoid, gray-white matter, and overall volumes (4 features); network properties (26 features); and volume and network properties (30 features). While the AUC for volumes and network metrics is lower than that for volumes only, the inflection point of the model associated with volumes and network metrics has consistently higher performance over 100 iterations using 5-fold validation. Each iteration uses a new generator for the 5-folds.