Experiment | Dataset | Size | GT | Seg. Method |
---|
A | Hammersmith | 100 | Yes | RF |
B | UKBB-2964 | 4805 | Yes | RF and CNN |
C | UKBB-18545 | 7250 | No | Multi-Atlas |
- Experiment A uses data from an internal dataset which is segmented with a multi-atlas segmentation approach and manually validated by experts at Hammersmith Hospital, London. These manual validations are counted as ‘ground truth’ (GT) and 100 of them are taken for the reference set used in all experiments. UKBB datasets are shown with their application numbers. In experiment C we segment with both random forests (RF) and a convolutional neural network (CNN). In C the CNN from Bai [4] is used