LIVER TUMOR ASSESSMENT WITH DCE-MRI Liliana Caldeira and Jo˜ ao Sanches Instituto de Sistemas e Rob ´ otica / Instituto Superior T´ ecnico 1049-Lisbon, Portugal ABSTRACT Dynamic-Contrast Enhanced MRI (DCE-MRI) is used in clinical practice to assess liver tumor malignancy. An algorithm to get infor- mation for automatic classification of tumors is presented. The Max- imum value and WashIn and WashOut rates, obtained from the per- fusion curves measured from the DCE-MRI images, are used in the classification process. The perfusion curves are described by a linear discrete pharmacokinetic (PK) model, based on multi-compartment paradigm where the input is the bolus injection. The arterial input function (AIF) that is usually estimated in the closest artery is as- sumed here to be the response of a second order linear system to the bolus injection. Therefore, the complete chain is modeled as a third order system with a single zero. The alignment procedure is performed by using the Mutual In- formation (MI) criterion with a non-rigid transformation to compen- sate the displacements occurred during the acquisition process. It is shown that the Maximum values and the WashIn and WashOut rates of the perfusion curves in malignant tumors are higher than in healthy tissues. This fact is used to classify them. Furthermore, it is also shown, that inside the tumor, the parameters associated with the perfusion curves for each pixel (time courses) present a higher variance than in the healthy tissues, which may also be used to increase the accuracy of the classifier. Examples using real data are presented. Index Terms— DCE-MRI, Pharmacokinetic Model, Perfusion Curve, Registration 1. INTRODUCTION Dynamic-Contrast Enhanced MRI (DCE-MRI) is used in clinical practice to get information about the malignancy of tumors. Ma- lignant tumors are known to have an active angiogenesis around them. So wider, more permeable and higher number of vessels can be found around malignant tumors when compared with healthy tis- sue or non malignant ones. In this case an increased contrast agent uptake is observed. Therefore, the revealed contrast kinetics param- eters may be used to characterize the tumors as malignant or be- nign [1]. Malignant tissues generally have an earlier contrast up- take, with rapid and large increasing when compared with benign tissues, which in general show a slower uptake. Cancer demonstrate rapid and high amplitude agent uptake, meaning large WashIn, fol- lowed by relatively rapid decreasing agent concentration, meaning large WashOut, while benign or normal tissue have smaller WashIn and WashOut. The maximum of the uptake is also higher in malign tumors than in benign. Therefore, the estimation of the perfusion curves may be used to classify the tumors. Correspondent author: Liliana Caldeira (llcaldeira@gmail.com).This work was supported by Funda˜ ao para a Ciˆ encia e a Tecnologia (ISR/IST plurianual funding) through the POS Conhecimento Program which includes FEDER funds. The traditional procedure to classify liver tumors uses liver func- tion tests and liver biopsy. This last one is an invasive procedure presenting the risk of spreading the cancer along the biopsy needle pathway. DCE-MRI is the preferred technique to assess tumor vascular characteristics because it is non invasive. However this is usually computationally intensive due to the huge amount of data generated by the MRI equipment which make them not appropriated in clinical practice. The processing time may be reduced by decreasing the volume of the ROI containing the tumor in the DCE-MRI data without de- creasing the spatial resolution, as shown in Fig. 1). A small ROI with small temporal resolution dataset is used in this work. This reduc- tion speeds up the alignment and analysis algorithms but increases the difficulty in the registration because less detail landmarks are available. Fig. 1. Selection of a ROI The signal intensity profile enhancement, after and before the contrast administration, along the time is used to estimate the perfu- sion curves. There are two ways to quantify perfusion. The first is based on the analysis of signal intensity changes, called tissue relax- ivity or semiquantitative. The second is based on the contrast agent concentration change using pharmacokinetics (PK) models. Semi- quantitative are straight forward to calculate but it is not completely supported on physiological reasons. PK models are therefore pre- ferred [2]. However, first, the patient motion occurred during the acqui- sition due to respiratory and cardiac activity must be compensated [3, 4]. In clinical practice, the evaluation of tumor is done mostly by human observation specially in the liver where several types of le- sions can occur. Highly vascularized tumors, having several arteries around them, are very visible in the arterial phase, during the first 30 seconds after injection [5]. The arterial phase is the most impor- tant to assess the malignancy and therefore is the phase that experts observe with more detail. The ultimate goal of this to develop an automatic tool to help the medical doctor in the diagnosis of liver tumors. In this paper, a PK model is estimated from the observed inten- sity profiles in a Statistical framework in order to deal with the noise 804 978-1-4244-2003-2/08/$25.00 ©2008 IEEE ISBI 2008