Computerized planning of liver surgery—an overview

Computerized planning of liver surgery—an overview

Computers & Graphics 26 (2002) 569–576 . Computerized planning of liver surgery—an overview Hans-Peter Meinzer*, Matthias Thorn, Carlos E. Ca! rdena...

369KB Sizes 2 Downloads 99 Views

Computers & Graphics 26 (2002) 569–576

.

Computerized planning of liver surgery—an overview Hans-Peter Meinzer*, Matthias Thorn, Carlos E. Ca! rdenas Department of Medical and Biological Informatics, Deutsches Krebsforschungszentrum, Im Neuenheimer Feld 280, 69120 Heidelberg, Germany

Abstract Liver surgery is a field in which computer-based operation planning has an enormous impact on the selection of therapeutic strategy. Based on pre-operative analysis of image data, liver operation planning provides a individual impression of tumor location, the exact structure of the vascular system and an identification of liver segments and subsegments. In this paper we present an operation planning system that is based on an object-oriented framework. This framework offers extensive automation of the integration process for software modules developed for medical software systems. The operation planning system can calculate the operation proposal results using two different approaches. The first method is based on the Couinaud’s classification system, which uses the main stems of the portal and venous trees. The second approach is a portal vein based method. These two approaches were compared using 23 liver CT scans. The volumetric data for individual segments demonstrates differences between the two segment classification methods. r 2002 Elsevier Science Ltd. All rights reserved. Keywords: Computer-based surgery; Computer tomography; Liver resection; Operation planning

1. Introduction A specific planning procedure for tumor resection is an important issue because of enormous individual anatomical variability in oncological liver surgery. Based on pre-operative diagnostics, a three-dimensional visualization and individualized operation planning for patients can be calculated using computer-aided methods during the operation planning phase. The planning procedure results in a proposal that includes the tumor location, the course of the intrahepatic vessels, the liver tissue to be resected and the remaining liver volume. These values can be interpreted as an approximation of liver functionality. Position and size of these volumes within the segmental structure provides an estimate of the patient’s operability. Planning must be based on the individual anatomy as defined by the structure of the intrahepatic vessel *Corresponding author. E-mail address: [email protected] (H.-P. Meinzer). URL: http://mbi.dkfz-heidelberg.de.

system. In the early 60s Couinaud proposed an anatomical liver segment model that sub-divided the liver into eight segments [1]. This classification is oriented along the large intrahepatic vessels. However, this is only an approximation of individually differentiated segment anatomy. A flexible and individually differentiated segment anatomy could not be established before the operation and could not be included in preoperative planning. Different approaches have been made for the planning [2] or the training and education [3] in this field of research. Using the computer-based operation planning system developed in Heidelberg it is possible to pre-operatively and non-invasively analyze individual anatomy [4]. The image data consists of contrasting agent enhanced CT images. Segmental classification is calculated by means of an analysis of portal vein structure. It defines volumes of the liver tissue components that are assigned to the different branches of the vessel tree. Also, tissue regions that are dependent on branches located within the safety margin are identified because these volumes need to be resected as well.

0097-8493/02/$ - see front matter r 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 0 9 7 - 8 4 9 3 ( 0 2 ) 0 0 1 0 2 - 4

570

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

The integration of computer-aided operation planning in this field requires that all steps of the analysis process are embedded in a framework that enables the reception of image data and forwarding of results. The individual procedural steps are: 1. segmentation of liver, tumor and other organs to be visualized in the operation planning proposal. These organs are referred to as the anatomical landmarks, 2. segmentation of the vessel system, 3. differentiation of the vessel tree, 4. conventional segment classification. This is an optional step for data sets with low resolution or insufficient vessel contrast, 5. calculation of a resection proposal and visualization of the results. The system receives the image data with the aid of the CHILIs radiological system [5]. CHILIs is capable of communicating with the imaging device and stores all data in its own patient and image database. Furthermore, it encrypts forwarded and stored data sets.

2. Implementation of the operation planning software The operation planning system is based on an objectoriented framework realized in C++. The framework consists of functional units, often partially finished, that can be extended or adapted to new needs by means of object-oriented technology. The reason for using this technology was the need for extensions with little overhead. We implemented the framework as an extension of the Model View Controller paradigm [6]. The framework enables the integration of dedicated modules into a software environment used to create diverse medical systems. Modules for segmentation and visualization are integrated into the framework as base components that are easily used and extended. In order to reduce integration expenditures for new elements, the framework offers generators for easy integration. Communication flow between components as well as between each module and a higher level architecture is a responsibility of the framework. We used the resulting framework to develop many systems including this operation planning system [7]. The implemented operation planning system consists of five modules belonging to each of the steps in the mentioned analysis procedure. 2.1. Segmentation of the anatomical landmarks The first step is the segmentation of the image data in order to tag the liver and areas of diseased liver tissue.

Other anatomical landmarks can also be identified for better orientation in the visualization. Various manual and semi-automatic algorithms used for this purpose were integrated into the segmentation model and are stable for use in the clinical routine. This module provides basic interactions such as region growing [8], merge, cut, threshold, active contours [9], undo, propagation [10,11], etc. [12–14]. The framework provides the opportunity for generating a graphical user interface for the image processing function. Therefore, new functions are easy to introduce. These algorithms can be used as two-dimensional tools to identify landmarks in each slice. To reduce the segmentation procedure time, many can also be applied to the whole volume. In order to facilitate the interaction between the end user and the segmentation module (Fig. 1), a set of interaction patterns was incorporated and standard parameter values was used. Interaction patterns determine the ways in which the parameter values of an algorithm are obtained. For example, they can be obtained from the input image with a mouse click. The application developers decided during the integration of the image-processing algorithm which interaction pattern will be used when the end user actives the algorithm into the framework.

2.2. Segmentation of the vessel tree A module divided into two parts performs vessel segmentation. The first one shows the input data and segmented landmarks (Fig. 2). The user can interactively change the image’s level/window values until only vessels are displayed. This sets a gray value. The portal vein system is used to calculate the resection strategy by selecting a starting point in the portal vein system’s stem. The gray value and the starting point define the parameters of the second module part. A modified algorithm for vessel tree location according to Zahlten [15] generates a symbolic description of the vessel tree. The algorithm then passes through the entire data set searching for connected vessel structures. A threedimensional reconstruction of the resulting vessel structure is shown which the user can rotate and zoom to analyze the result [16]. If the result does not correspond to the expected vein structure, the user can set another starting point or new range values. The calculation of the resection strategy is currently based on the portal system structure but both parts of the venous system are enhanced. Because of noise and low resolution, pseudo-connections may occur between the portal and hepatic systems and, therefore, parts of the hepatic system may be included in the segmentation result. Usually, an editing step becomes necessary to separate portal from hepatic veins.

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

571

Fig. 1. Module for the semi-automatic segmentation of the liver- and tumor tissue.

Fig. 2. Segmentation of the vessels. Setting a seed point into the two-dimensional view on the left leads to an three-dimensional visualization of the segmented vessels on the right.

2.3. Separating vessel systems The main component of this module is the threedimensional reconstruction created from the preceding module. In this module the user can edit the vessel system (Fig. 3). The vessel branch that contains the starting point of the segmentation algorithm is colored green. The segmentation process may have generated

invalid connections between the two venous systems but these can be severed interactively [16,18]. The location of the invalid connections can be detected by calculating the path from the portal stem to a part of the hepatic system that belongs to the segmentation result. In this display the location where the segmentation result must be severed is easily detected. Each branch can be chosen interactively

572

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

Fig. 3. Separating the two vessel systems (vena porta and liver veins) with five clicks on average.

and tagged as a ‘‘stop branch’’. This tag severs the pseudo-connection. Only those parts of the vessel tree that are still connected to the portal stem will be brightly displayed. This is done until all parts of the hepatic system are removed from the segmentation result.

2.4. Conventional segment classification Contrast enhanced vessel imaging is not possible for every patient. In addition to the vessel-based segment classification the operation planning system is able to calculate a resection proposal for data sets with low contrast enhancement for the vessel system. In these cases, two components were integrated in the operation system that allow interactive positioning of segment interfaces in a three-dimensional view based on the segmented organs. This segment classification conforms to the model proposed by Couinaud. It consists of four planes (equivalent to the three main liver trees and the primary portal vessel bifurcation) in the three-dimensional liver model, which divide the liver into eight segments [17]. The first component is used to initialize the points that define plane positions by fixing points in the main stems of the intrahepatic veins (Fig. 4). The second component shows the segmented organs and the planes that divide the liver. The user can manipulate these planes by moving the points defined in the first module (Fig. 5).

2.5. Resection planning and visualization The last step of the planning procedure is the presentation of the results. This step includes a visualization that shows the tumor position in relation to the vessels. Vessels located inside the safety margin can also be shown. Operation planning will visualize the proposed resection lines on the surface of the liver (Fig. 6). The quantitative results include total liver volume and tumor size. The volume of the sound liver tissue that must be resected in addition to the tumor is particularly important because its size approximates the loss of liver functionality. An extension of the operation system was implemented to determine the difference between the segment branches of the portal vessel stem produced by vessel tree classification and the segment area as calculated with Coinaud’s classification method (Fig. 7). The subsequent vessel tree separation was manually assigned to the segments they service. This was done by clicking on the respective branches of the three-dimensional visual graphic description produced from the vessel stem in the third module. In a randomized study in cooperation with the Department of Surgery at Heidelberg University, the CT data of 64 patients was analyzed. Of these patient data sets, 23 conformed to the criteria and could be used in a segment comparison. An evaluation of the relative location of equivalent segments to each other was made possible by a calculation of common voxels in the two methods

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

573

Fig. 4. Module for the initialization of the classic Couinaud-Model. Landmarks have to be set into the liver veins, vena porta, and vena cava with the help of the two-dimensional view.

Fig. 5. The classic Couinaud-Model.

compared. We calculated k values for individual segments and surgically relevant interventions. These parameters enable quantitative evaluation of two different random samples as a function of their location and form. The k value calculates the agreement of one voxel group with another group of other volume elements.

Depending on the form and shape of the hepatic segments the two methods will yield significantly different results (Fig. 8). Only the two hemilivers divided by the middle hepatic vein and the left liver lobe marked by the falciform ligament as their respective anatomic landmarks show very good agreement between the classification systems

574

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

Fig. 6. Visualization of the planning results (two vessel systems, tumor, resection proposal in compliance with the security margin).

Fig. 7. Visualization of the results. The difference of the two segment classification procedures can be seen.

(k>0.7). With respect to specific liver segments, we found poor correlation (ko0.4) for four segments and good correlation (ko0.7) for four other segments between the two different classification methods. Furthermore, portal segmentation of the human liver appears to be independent of hepatic vein architecture.

3. Discussion The high variability of liver anatomy makes liver resection planning a useful application for computeraided planning procedures. We developed a software module capable of segmenting the liver, tumor and intrahepatic vessel structures. The objective of the

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

575

Fig. 8. Comparison of the segment classification of Couinaud (white lines) and the portal vein based classification (colored blocks).

planning procedure is an individualized calculation of liver sub-segment classification that will localize the interfaces between different sub-segments. A resection strategy based on sub-segment interfaces as well as estimating post-operative liver volume is proposed using this information. With this system, a pre-operative statement can be made for the first time with respect to size, shape and location of liver segments and their relative share in overall liver volume. In addition, it is possible to image the three-dimensional vascular structure that defines dependent areas.

[6] [7]

[8]

[9]

[10]

References [1] Couinaud C. Le foie, e! tudes anatomiques et chirurgicales. Paris: Masson, 1957. [2] Selle D, Spindler W, Schenk A. Computerized models minimize surgical risk. Diagnostic Imaging Europe 2000;16(9):16–20. [3] Marescaux J, Clement J-M, Tassetti V, et al. Virtual reality applied to hepatic surgery simulation: the next revolution. Annals of Surgery 1998;228(5):627–37. . [4] Glombitza G, Lamade W, Demiris AM, Gopfert MR, Mayer A, Bahner ML, Meinzer HP, Richter G, Lehner Th, Herfarth C. Virtual planning of liver resections: image processing, visualization and volumetric evaluation. International Journal of Medical Informatics 1999; 53(2–3):225. [5] Engelmann U, Ells.asser C, Schwach M, Soellig C. Evaluation of the CHILI teleradiology network after 3 years of clinical routine. In: Gell G, Holzinger A, Wiltgen M, editors. From PACS to internet/intranet, informationsystems, multimedia and telemedicine. Proceedings Euro-

[11]

[12]

[13]

[14]

[15]

[16]

PACS 2000, Wien, Oesterreichische Computergesellschaft, 2000. p. 104–10. Goldberg A. Smalltalk-80. The language. MA, USA: Addison-Wesley, 1989. Cardenas CE. Konzeption und Implementierung eines objektorientierten Frameworks und dessen Komponenten fuer die multimodale praeoperative Operationsplanung in der Leber- und Nierenchirurgie. Ph.D. thesis, University of Heidelberg, Medical Faculty, 2001. Justice RK, Stokely EM, Strobel JS, Ideker RE, Smith WM. Medical image segmentation using 3-D seeded region growing. Proceedings of the SPIE Image Processing 1997;3034:900–10. Kunert T, Heiland M, Meinzer HP. User-driven segmentation approach: interactive snakes. Proceedings of SPIE Medical Imaging, Image Processing, 2002, in press. Herman GT, Zheng J, Buchotz CA. Shape-based Interpolation. IEEE Computer Graphics & Applications 1992;12(3):69–79. Raza SP, Udupa J. ShapeBased interpolation of multidimensional objects. IEEE Transaction on Medical Imaging 1990;9(1):32–42. Olabarriaga SD, Smeulders AWM. Interaction in the segmentation of medical images: a survey. Medical Image Analysis 2001;5:127–42. Pham DL, Xu C, Prince JL. A survey of current methods in medical image segmentation. Annual Review of Biomedical Engineering, Palo Alto, 2000. p. 315–337. Haralick RM, Shapiro LG. Image segmentation techniques. Computer Vision, Graphics, and Image Processing 1985;29:100–32. Zahlten C, Juergens H, Peitgen HO. Reconstruction of branching blood vessels from CT-data. In: Goebel M, Muller . H, Urban B, editors. Visualization in scientific computing. Wien: Springer, 1995. p. 41–52. Thorn M, Vetter M, Cardenas C, Hassenpflug P, Fischer L, Grenacher L, Richter GM, Lamade W, Meinzer HP. Interaktives Trennen von Gef.ab.aumen am Beispiel der

576

H.-P. Meinzer et al. / Computers & Graphics 26 (2002) 569–576

Leber. In: Lehmann T et al., editors. Informatik Aktuell— Bildverarbeitung f.ur die Medizin 2001—Algorithmen, Systeme, Anwendungen. Heidelberg: Springer, 2001. p. 147–51. [17] Thorn M, Sonntag S, Glombitza G, Lamad!e W, Meinzer HP. Ein interaktives Tool fur . die Segmenteinteilung der Leber in der chirurgischen Operationsplanung. In: Evers H, Glombitza G, Lehmann T, Meinzer HP, editors.

Informatik Aktuell—Bildverarbeitung fur . die Medizin 1999—Algorithmen, Systeme, Anwendungen. Berlin, Heidelberg, New York: Springer, 1999. p. 155–9. [18] Kunert T, Thorn M, Meinzer HP. Visualization and attributation of vascular structures for diagnostics and therapy planning. Proceedings of Medicine Meets Virtual Reality (MMVR 2002), 2002, in press.