FIGURE 4.

Training progression according to training iterations and training networks. (A) Qualitative training progression for the MRI of Patient 1. Upper two rows: training progression images using the FCN-32s coarse segmentation network. Lowest row: training continuation using FCN-8s, which showed the fine segmentation network. Segmentation using FCN-8s has smoother and clearer margins than that using FCN-32s only. (B) Training progression for MRI of Patient 2. (C) Quantitative improvements in semantic segmentation accuracy in terms of mean intersection over the union (IoU) values according to the number of training iterations and the amount of data used. Validations were performed in 360 augmented images for all graphs. The numbers of iterations are displayed on a log scale for better visualization of improvement curves.