Magnetic Resonance Imaging, Vol. 14, No. I, pp. 73-92, 1996 Copyright 0 1996 Elsevier Science Inc. Printed in the USA. All rights reserved 0730-725X/96 $15.00 + .OO ELSEVIER l Original Contribution 0730-725X(95)02040-3 SIMULATION OF MRI CLUSTER PLOTS AND APPLICATION TO NEUROLOGICAL SEGMENTATION ANDREW SIMMONS,*? SIMON R. ARRIDGE,$ GARETH J. BARKER,$ AND STEVEN C.R. WILLIAMS* *Department of Neurology, Institute of Psychiatry, De Crespigny Park, London, SE5 8AZ, UK, tDepartment of Medical Physics, Kent and Canterbury Hospital, Canterbury, Kent, CT1 3NG, SDepartment of Computer Science, University College London, Gower Street, London WClE 6BT, UK, and $NMR Research Group, Institute of Neurology, Queen Square, London WCIN 3BG, UK The advent of magnetic resonance imaging has provided new opportunities for volume measurement of tissues, with applications increasing dramatically in recent years. Cluster classification techniques have proved the most popular for volume measurement, yet little attention has been paid to how the choice of images for analysisaffects the quality and ease of segmentation. To address this issue, we have developed a system to simulate MRI cluster plots using multicompartmental anthropomorphic software models of anatomy, and components for image con- trast, signal-to-noise ratio, image nonuniformity, tissue heterogeneity, imager field strength, the partial volume effect, correlation between proton density, T1and T2, and a variety of data preprocessing techniques. The effect of these components on tissue cluster size, shape, orientation, and separation is demonstrated. The simulation allows an informed choice of pulse sequence, acquisition parameters, and data preprocessing for cluster classifi- cation to be made as well as providing an aid to interpretation of acquired data cluster plots and a valuable educa- tional tool. The systemhas been used to choose suitable images for neurological segmentation of grey matter, white matter, CSF, and multiple sclerosis lesions using spin-echo, inversion recovery, and gradient-echo pulse sequences. Constraints on image selection are discussed. Keywords: Magnetic resonance imaging; MRI; Image processing; CIu.%er classification: Simulation. INTRODUCTION Segmentation of MRI data has become increasingly popular in recent years for a variety of tasks including volume measurement, 3D visualisation of data, and im- age feature analysis. Interactive outlining of regions of anatomy, pathology, or function is time consuming and particularly prone to inter- and intraobserver variabil- ity. For this reason a variety of semiautomated and au- tomated approaches to MRI segmentation have been developed. By far the most prevalent have been meth- ods based upon cluster classification.‘-” Such meth- ods take advantage of the intrinsically multiparametric nature of MRI by utilising more than one registered im- age of the same region of anatomy for segmentation. These techniques can deliver an improvement in the quality of segmentation over algorithms applied only to a single image. In its most simple form a cluster plot is a two-dimensional (2D) intensity histogram where the two axes represent intensity for the two base images. This concept is easily extended to three or more dimen- sions as required. By defining regions within the clus- ter plot using automated or semiautomated techniques it is possible to identify regions in the base images that correspond to particular tissue types. High-quality data segmentation can best be achieved by concentrating on the three steps of data acquisition, preprocessing, and image analysis. Despite the popu- larity of cluster classification techniques, little attention has been paid to data acquisition, that is, the optimal choice of base images for cluster classification, and the effect that this has on the quality and ease of segmen- tation. Appropriate choice of pulse sequences(s) and acquisition parameters is a vital part of the cluster clas- RECEIVED 12/ 19194; ACCEPTED 7/12/95. ment of Neurology, Institute of Psychiatry, De Crespigny Address correspondenceto Dr. Andrew Simmons, Depart- Park, London SE5 8AZ, UK. 73