Computer Vision News - July 2021

3 Summary 7 To estimate the optimal parameters of the network and perform the automatic (deep learning-based) evaluations, a cross-parameter grid search was performed, by varying hyper-parameters such as the network depth, number of epochs and kernels. There was no overfitting for the optimal model, but some fluctuations in performance for the fourth validation fold were seen. Figure 3: Here is the three-dimensional rendering of the manual segmentation trials from the two readers for three chosen subjects. Figure 4: Oblique coronal plane with a super-imposed automatic segmentation using three different thresholds (Qu + 1.5 IQR, Qu + 2.5 IQR, maximum threshold). Towards Deep Learning- assisted Quantification ...