Automatic MR imaging display window adjustment with a neural network

This paper reports on a system developed to adjust automatically the MR imaging display window width and level (WWL) with a neural network. Six features of MR images are extracted with use of the gray-level histogram of various MR images of the brain. A neural network learns the relationship between six features and the optimum display WWL by use of training images. The learned neural network calculates the index showing the goodness of a displayed image. An adequate display WWL can be determined effectively using the hill climbing method: The maximum index is searched from roughly sampled WWLs, the WWL sampling intervals are gradually made finer, and the WWL with maximum index searched in b is selected as a most adequate display WWL