H. Schöller (1,2) , M. Blaickner (3), W. Hofmann (4) , H. Deutschmann (1,2) , M. Kopp (1) , K. Wurstbauer (1) , F. Sedlmayer (1,2)


Combined 3D Segmentation of PET- and CT-Datasets

concerning Dosimetry, Tumour Staging and Treatment Management in Targeted Radionuclide Therapy and External Beam Radiotherapy


1) University Clinic for Radiotherapy and Radio-Oncology, Paracelsus Medical University Salzburg,

 2) radART – Institute for research and development on Advanced Radiation Technologies at the Paracelsus Medical University Salzburg,

 3) Medical Physics Department, Radiation Safety and Applications Seibersdorf, Austrian Research Centers GmbH – ARC,

4) Division of Physics and Biophysics, Department of Materials Engineering and Physics, University of Salzburg.


Textfeld: Introduction


Volume segmentation is a wide-spread method to identify functional and anatomical entities on volume datasets of various imaging devices as MRI, CT, PET, or US. Diagnosis, planning, the treatment and its evaluation are topics based on the combined analysis of both anatomical (e.g. CT or MRI) and functional (PET, SPECT) data. Segmentation aims in defining anatomical structures and tumours. Precise estimation of their volumes is crucial for the accuracy of calculated absorbed doses, as well for External Beam Radiotherapy (EBRT), as for Targeted Radionuclide Therapy (TRT).

Textfeld: Application I:  External Beam Radiotherapy (EBRT) – Staging, Tumour Volume Definition 
and Treatment-Evaluation


In TRT the situation is somehow different compared to EBRT. Calculation of absorbed dose is not only based on 3D-electron-density-distribution, but also on temporal and spatial distributions of administered activities which are monitored in dynamic PET-studies. Analysis of the characteristics provides diagnostic insight and is supportive for planning and dosimetry of TRT. The treatment itself is performed by administering the same tracer, now coupled to a

Beta Minus emitting component. As TRT is an inherently multidisciplinary approach and volume segmentation is a shared central task for TRT- and EBRT, this project is designed as collaboration of multiple institutes.

Positron emission tomography (PET), mainly using [18F]fluoro-Deoxyglucose (FDG) as tracer, is widely accepted as diagnostic tool in oncology due to itsability to provide functional information by extending spatial information from diagnostic- or planning– CT.



Figures 1 and 2: Ecat-Exact HR and Biograph 6 PET Scanner, Siemens Medical, Malvern (US)

Erlangen (D)

Staging: Clinical assessments of PET and PET/CT showed differences, particulary in cases of NSCLC [Baar06], in both staging and treatment management, that influence clinical target volumes to a considerable extent.

Planning and Evaluation of Treatment Response:

Examinations about the influence of PET/CT on treatment-planning, target volume delineation and evaluation of EBRT have been performed for the sites Head and Neck, Brain, Lung, Lymphoma, Gynaecology and Rectum.

One of the latest clinical studies evaluated the feasibility to correlate intratumour-heterogeneity as visualized of 18F-FDG PET with histology for NSCLC. It is suitable for correlating intra-tumour heterogeneity in tracer uptake with histology [Baar08] .



              Application II: Dose-Calculation for Targeted Radionuclide Therapy (TRT)




In Targeted radionuclide therapy (TRT) unsealed radioactive compounds, consisting of a ‘carrier’-component and an attached radionuclide enrich in tumour tissue due to the metabolic properties of the carrier and its kinetics.


Figure 3: Rendered view of absorbed dose distribution from SPECT/CT
segmentation of I-131 mIBG therapy of neuroblastoma [Flux06]


Segmentation of combined PET and CT is a prerequisite for treatment planning in TRT. By  using positron- and electron-emitting radionuclides of the same element attached to the same tracer the Beta Plus component (e.g. I-124) is suitable for visualization of the agents biodistribution via PET and the Beta Minus component (e.g. I-131) can be administered for therapy since it experiences the same kinetics.

To implement patient-specific dosimetry for targeted radiotherapy, calculations have to be performed based on spatial and temporal distribution of the radioactive agent and established electron-densitydistribution in the nearest neighbourhood of spots with relevant


The Medical Physics Department, Radiation Safety and Applications Seibersdorf of the Austrian Research Centers GmbH develops new and deterministic calculation algorithms which are validated by Monte Carlo radiation transport simulations. Calculation of the absorbed dose rate at each target-point is performed by convolution of specific cumulated activity at each source-point with a three-dimensional pointsource-kernel for the radiation[Rein07].

The current project aims to perform segmentation of clinical data in order to provide the input for dose calculations in TRT.

Methods for Automatic Segmentation





A main shortcoming of PET is that exact tumour-borders are not well defined, making visual delineation error-prone. Available software is mainly restricted to threshold-based-strategies, calculating either thresholds relative to maximum or related to source-to-background-ratios. Combining processing of co-registered CT and PET and coordination of specialized analysing-methods of different modalities can enhance usability of PET/CT.

Although documented segmentation-algorithms offer a great variety of methods, clinical implementations of automatic PET-related methods are based on mainly two approaches: The majority are (adaptive) threshold-oriented and only a minority uses gradient-oriented procedures.

Algorithms for automatic segmentation can be classified into three categories:

• structural techniques , based on local structural information

• statistical (stochastic) techniques

• hybrid techniques, combining structural and stochastic methods

Other common classification systems divide segmentation methods in

• Point-oriented methods

• Edge-detection based strategies

• Region-based procedures

• Texture-based trials

• Model-based strategies and object-recognition

• Atlas-guided approaches

Our specific approach combines PET- and CT-oriented techniques merging the CT’s high spatial resolution and the PETs functional information.

Overlay-techniques, and volume-projections in „Beams-Eye-View“ are available through our inhouse developed ROKIS-Software RT² and allow for proper evaluation and clinical assessment of new developed algorithms.


First Results from Applying Thresholds at PET-Data




Building PET-volumes by applying absolute thresholds allow for individual analysis of PET/CT-datasets. Relative thresholds of about 40% of the maximum SUV-values or maximum activity-levels are proper settings for segmenting true tumour-volumes [Erdi95], [Erdi97], [Cier05], [Vall93], [Davi02], [Bent04]. Overlay-functions give spatial orientation in combining PET-findings with anatomical information. Threshold-based volumes can be projected by Maximum-Intensity-Projection (MIP) in various treatment-field-geometries and visualized in Beams-Eye-View. Examples shown in figs. 4 and 5 are concerning bronchus carcinoma and gynaecological findings.

Figure 4: Threshold-limited Maximum-Intensity-Projection (MIP) of PET-Study of bronchus carcinoma of the medial lobe T2 N2 M0 and corresponding GTV/PTVs: A: ventral field, B: corresponding left view and C: Overlay with Sum-Projection of co-registered CT-dataset.



Figure 5: Threshold-limited MIP of PET-Study of Gynaecological finding A: left view, B: Corresponding ventral view , C: Corresponding ventral overlay with threshold-filtered sum-projection of co-registered CT-dataset and field geometry.



[Bent04]: S.M. Bentzen, High-tech in radiation oncology: should there be a ceiling? Int J Radiat Oncol Biol Phys 2004;58:320–30, [Baar06]:  A.v. Baardwijk, B.G. Baumert, G. Bosmans, M.v. Kroonenburgh, S. Stroobants, V. Grégoire, Ph. Lambin, D.De Ruisscher, The current status of FDG-PET in tumour volume definition in radiotherapy treatment planning, Cancer Treatment Reviews (2006) 32, 245-260, [Cier05]: I. F. Ciernik, M. Huser, C. Burger, J. Bernard Davis, G. Szekely, Automated functional image-guided Radiation Treatment Planning for rectal cancer, Int. J. Radiation Oncology Biol. Phys., Vol. 62, No 3, 893-900, 2005, [Baar08]: A. v. Baardwijk, G. Bosmans, R.J.v. Suylen, M.v. Kroonenburgh, M. Hochstenbag, G. Geskes, P. Lambin, D.de Ruysscher, Correlation of intra-tumour heterogeneity on 18F-FDG PET with pathologic features in non-small cell lung cancer: A feasibility study, Radiotherapy and Oncology 87 (2008) 55-58, [Davi02]:  J. B. Davis, B. Reiner, A. Dusserre, J.Y. Giraud, M. Boll, Quality assurance of the EORTC trial 22911. A phase III study of post-operative external radiotherapy in pathological stage T3N0 prostatic carcinoma: the dummy run, Radiotherapy and Oncology 64 (2002) 65–73, [Erdi95]: Y. E. Erdi, Barry W. Wessels, Murray H. Loew, and Alev K. Erdi, Threshold Estimation in Single Photon Emission Computed Tomography and Planar Imaging for Clinical Radioimmunotherapy, CANCERRESEARCH(SUPPL.)55. 5823s.-5826s, December 1995, [Erdi97]:   Y. E. Erdi, O. Mawlawi, Steven M. Larson, M. Imbriaco, H. Yeung, R. Finn, John L. Humm, Segmentation of Lung Lesion Volume by Adaptive Positron Emission Tomography Image Thresholding, Cancer, Volume 80 Issue S12 , Pages 2343 - 2753 (15 December 1997), [Krem07]: R. Kremslehner, Computer-Assisted Localization of Mice Organs in Micro Positron Emission Tomography, Diplomarbeit am Institut für Festkörperphysik der Technischen Universität Wien und Austrian Research Centers, Seibersdorf, Wien 2007. [Rein07]: D. Reiner, Deterministic algorithms for the calculation of tumour and organ doses in targeted radionuclide therapy and their implementation with voxel models, unveröffentlicht, Austrian Research Centers Gmbh., Seibersdorf 2007, [Vall93]: J.F.Valley, M. Ro, Comparison of treatment techniques for lung cancer. Radiother Oncol, 1993;28:168–73, [Flux06]: G. Flux et al, The Impact of PET and SPECT on Dosimetry for Targeted Radionuclide Therapy. Z. Med Phys. 16 (2006) 47-59