Dynamic contrast-enhanced (DCE) CT (or perfusion CT) provides a non-invasive method for assessing tumour blood flow (1). DCE CT involves the intravenous injection of an iodinated contrast medium (tracer) followed by sequential imaging to monitor the traversal of tracer within the tissue of interest. Tumour enhancement curves derived from DCE CT reflect microcirculatory states of the underlying neovasculature and are commonly analysed by tracer kinetic modelling (2). In this study, an automatic method for classification of tumour voxel-level enhancement patterns is introduced for monitoring change...

Automatic classification of voxel-level enhancement patterns was performed using a combined principal component analysis (PCA) and Kmeans clustering method. PCA was used as a dimension reduction step to determine the appropriate number of distinct enhancement types (clusters) within the DCE CT dataset. Kmeans clustering is subsequently employed to classify the enhancement curve in each voxel into one of the clusters (3). To demonstrate clinical feasibility, the automatic classification method was used to analyse a patient study case with serial brain DCE CT scans performed before,...

Before and immediately after radiotherapy (Fig. 1), classification analysis revealed that the tumour voxels are largely associated with a distinct cluster, which exhibited a strong wash-in and plateau enhancement pattern, different from the surrounding tissues. Six weeks after radiotherapy (Fig. 2), tumour voxels are classified into another distinct cluster with a gentler wash-in and progressively increasing pattern. Sixteen weeks after radiotherapy (Fig. 3), voxel locations where the tumour previously resided are no longer associated with a separate cluster from the surrounding tissues, indicative of tissue...

A combined PCA-Kmeans clustering method can be used for automatic classification of tumour enhancement patterns on DCE CT to monitor change in tumour microcirculatory states with radiotherapy.

Tong San Koh (email: [email protected]), Choon Hua Thng, Patrick T.H Teo and James B.K. Khoo are with the Department of Oncologic Imaging, National Cancer Centre Singapore. Dennis L.H. Cheong is with theClinical Imaging Research Centre, A*STAR-NUS, Singapore. Keith H.C. Lim is with theDepartment of Radiation Oncology, National University Cancer Institute, Singapore. Tchoyoson C.C. Lim is with theDepartment of Neuroradiology, National Neuroscience Institute, Singapore.

1. Miles KA, Lee TY, Goh V, et al. Current status and guidelines for the assessment of tumour vascular support with dynamic contrast-enhanced computed tomography. Eur Radiol 2012; 22:1430-1441. 2. Koh TS, Bisdas S, Koh DM, Thng CH. Fundamentals of tracer kinetics for dynamic contrast-enhanced MRI. J Magn Reson Imag 2011; 34:1262-76. 3. Koh TS, Hartono S, Thng CH, et al. In vivo measurement of gadolinium diffusivity by dynamic contrast-enhanced MRI: A preclinical study of human xenografts. Magn Reson Med 2013;69:269-276. 4. Bisdas S, Foo...