<< Click to Display Table of Contents >> Navigation:  Case Studies - Application of PAI > Mouse Bone Trabecular Segmentation > Sample Preparation

In the first approach - rapid model - a subset of 70 slices from two femurs were imported into a database along with their reference segments. The microCT and segment series were associated using automatic association. A learning set was prepared with all samples and segmentation using uNET segmentation architecture. 60 samples were assigned for training - to be split into 48 training and 12 validation - and 10 samples were labeled as Test samples for evaluation of the model.

In the second approach - detailed model - all slices from the one femur with full reference segments and seven of the eight tibia were imported into a database along with their reference segments. The microCT and segment series were associated using automatic association. In total 3509 2D samples were available for training. A learning set was prepared with all samples and segmentation using uNET segmentation architecture. 3509 samples were assigned for training - to be split into 2486 training and 622 validation - and 401 were labeled as Test samples for evaluation of the model.