Bone tumor segmentation from MR perfusion images with neural networks using multi-scale pharmacokinetic features

Bone tumor segmentation and the distinction between viable and non-viable t umor tissue is required during the follow-up of chemotherapeutical treatmen t. Monitoring viable tumor area over time is important in the ongoing asses sment of the effect of preoperative chemotherapy. In this paper, features d erived from a pharmacokinetic model of tissue perfusion are investigated. A multi-scale analysis of the parametric perfusion images is applied to inco rporate contextual information. A feed-forward neural network is proposed t o classify pixels into viable, non-viable tumor, and healthy tissue. We ela borate on the design of a cascaded classifier and analyze the contribution of the different features to its performance. Multi-scale blurred versions of the parametric images together with a multi-scale formulation of the loc al image entropy turned out to be the most relevant features in distinguish ing the tissues of interest. We experimented with an architecture consistin g of cascaded neural networks to cope with uneven class distributions. The classification of each pixel was obtained by weighting the results of five bagged neural networks with either the mean or median rules. The experiment s indicate that both the mean and median rules perform equally well. (C) 20 01 Elsevier Science B.V. All rights reserved.