3-D Brain Tumor Segmentation Using Deep Learning

My question is reated to the example in the object. I tried to exactly replicate the example with any modification to the orignal script.

I downloaded the data, preprocessed it, created the graph and started the training process. The first step of the training process are displayed in the folowing fig.

I was wondering if things are going right. I would have expected a more slow accuracy increase. Is this correct? Are the iteration still too few?

I think the behavior is too different respect the images in the matlab "3-D Brain Tumor Segmentation Using Deep Learning" example.

Thanks