TAPIR - Tool for Automatic Planning In Radiotherapy

Figure 1: Architecture of the embedded KBS. Dependent on the definition of the target volume the system selects suitable skeletal plans from the knowledge-base. Once adapted to the given case the plans are optimized with prestored methods and presented to the planner for further evaluation and manual optimization.
© dkfz.de

TAPIR is a knowledge based System, which can generate automatically pre-optimised radiotherapy plans for a given patient based on already stored treatment descriptions. Since TAPIR allows the definition of rules, how stored treatment plans can be adjusted to individual patient geometries it can dynamically generate a series of treatment proposals.

Since TAPIR allows the storage of standard treatment techniques which can be applied automatically to new patient data sets it can speed up treatment planning in clinical routine. Dose distributions for standard beam configurations can be calculated automatically without any user interaction and therapists can use the proposals as starting point for their own optimisations.

The system’s architecture was described in detail in a contribution to the XII ICCR 1997 in Salt Lake City:

M.-A. Keller-Reichenbecher, M. van Kampen, R. Bendl, G. Sroka-Pérez, J. Debus, W. Schlegel

A Knowledge-Based System for Rapid Generation of Alternative Pre-Optimized Plans for Conformal Radiotherapy Planning

We present here major parts of this paper:


Figure 2: Schematic display of how skeletal plans can be constructed from a set of different components. Skeletal plans can inherit components from other predefined skeletal plans, resulting in a minimal redundancy as well as in a fast and elegant construction.
© dkfz.de

The designed system requires as input the predefined 3D contour definitions (contour outlines) of the target volume and organs at risk (OAR) as well as contiguous CT slices used by the dose calculation algorithm to generate a 3D dose distribution.

Each OAR and target volume localization is labeled with a unique key according to ICD-O (International Classification of Diseases for Oncology, 2nd edition, WHO, 1990). The KBS uses the keys for plan selection, plan generation, verification and evaluation of the results.

Knowledge representation with skeletal plans

The system is based on skeletal plans, each representing a clinical strategy for the irradiation of a specific target volume (here planning target volume PTV). It adapts one or more of the prestored skeletal plans to a given case and presents the calculated results (e.g., dose distributions, dose statistics, dose-volume histograms, biological and physical objectives) to the planner for further evaluation and manual optimization with conventional 3D radiotherapy planning tools (Fig.1).
Methods used during the generation of plans

A skeletal plan relies on the following methods:

  • Beam’s Eye View volumetry to find the best beam portals of a stored beam configuration as proposed by Chen et al. and Myrianthopoulos et al. [7, 8] and Bendl et al. [9].
  • Mathematical morphology techniques to calculate irregular field margins for each beam.
  • Simulation of the results with fast dose calculation.
  • Methods to evaluate the calculated dose distributions by biological models and dose statistics.

Figure 3: Dose-volume histogram of two plans for prostate cancer conformal treatment (4-field box, normalized to doses at isocenter). An example of an automatically generated plan compared to a manually optimized plan. 1 denotes the target volume (t), 2=colon (c), 3=bladder (b), 4=rectum wall (r), 5/6=right and left femoral head (f) of the automatically generated plan, 7-12 the target-volume and OAR of the manually generated plan, respectively.
© dkfz.de

Each skeletal plan can inherit information from a previous plan to facilitate the achievement of new concepts required by clinical routine (Fig.2).
Rapid plan generation

Rapid plan generation is utilized by using the parallel character of the skeletal plans. All tasks scheduled by the system (e.g., plan generation, dose calculation, evaluation) are executed on a "virtual" parallel system consisting of one or more UNIX workstations or even PCs running LINUX connected via network.
Plan evaluation

DVHs are useful to quickly compare two dose distributions but are usually difficult to be interpreted if employed for more than two generated plans.

Another way to evaluate 3D treatment plans is the use of physical or biological objectives. Dependent on the number of skeletal plans for a target localization the system can produce more than a dozen different plans for a clinical case. We use these objectives in our system as scoring functions and as a preselection criterion for suitable plans. This reduces the number of plans to be viewed by the planning team.

The biological objectives are based on TCP/NTCP calculations as published by Webb et al., Lyman and Kutcher et al. [10, 11, 12]. A summary score is calculated as follows:

The physical objective is calculated as a sum of quadratic dose differences of each OAR for doses higher than TD5/5 plus quadratic dose differences of the dose values in the target volume from the prescription dose.
Embedding of the system

The KBS is embedded into the VOXELPLAN system developed at DKFZ. The VIRTUOS module of VOXELPLAN provides additional options to simultaneously display and evaluate the generated results of alternative plans.

Because of its modular architecture, the system can be included into other available 3D radiotherapy planning software tools with minimal adaptation.

Figure 4: Biological objectives calculated for the target volume and four OAR of four plans for bronchial cancer treatment (TCP, NTCP and total score). The automatically generated plans were normalized to the TCP value of the manually generated plan by decreasing the initial prescribed dose of 60 Gy. The prescribed dose was applied to the 80% isodose of each dose distribution. All dose distributions were normalized to their dose at isocenter. A higher total score describes a "better" plan
© dkfz.de


gen. Plan

no. 1

no. 2

no. 3
















Figure 5: Physical objectives calculated for the target volume, and four OAR of the same plans as displayed in Fig. 4. Normalization was the same as described in Fig. 4. Lower values describe a better conformation to the target volume and lower doses in the OAR. A lower total score decribes a "better" plan. The beam configurations are listed in Table 1.
© dkfz.de


  • [1] Kalet, I.J. and Paluszynski, W. A Production Expert System for Radiation Therapy Planning. In: American Association for Medical Systems and Information - Proceedings, 1985, p. 315-319.
  • [2] Kalet, I.J. and Jacky, J.P. Knowledge-Based Computer Simulation for Radiation Therapy Planning. In: The Use of Computers in Radiation Therapy, edited by Bruinvis, I.A.D North-Holland: Elsevier Science Publishers, 1987, p. 553-559.
  • [3] Zink, S. The Promise of a New Technology: Knowledge-Based Systems in Radiation Oncology and Diagnostic Radiology. Comput.Med.Imaging Graph. 13:281-293, 1989.
  • [4] Kalet, I.J. and Paluszynski, W. Knowledge-Based Computer Systems for Radiotherapy Planning. Am.J.Clin.Oncol. 13 (4):344-351, 1990.
  • [5] Kalet, I.J., Jacky, J.P., Austin-Seymour, M., Hummel, S.M., Sullivan, K.J. and Unger, J.M. PRISM: A New Approach to Radiotherapy Planning Software. Int.J.Radiat.Oncol.Biol.Phys. 36 (2):451-461, 1996.
  • [6] Vijayakumar, S. and Chen, T.Y. Implementation of Three Dimensional Conformal Radiation Therapy: Prospects, Opportunities, and Challenges. Int.J.Radiat.Oncol.Biol.Phys. 33 (5):979-983, 1995.
  • [7] Chen, G.T.Y, Spelbring, D.R., Pelizzari, C.A., et al. The use of beam's eye view volumetrics in the selection of non-coplanar radiation portals. Int.J.Radiat.Oncol.Biol.Phys. 23:153-163, 1992.
  • [8] Myrianthopoulos, L.C., Chen, G.T.Y, Vijayakumar, S., Halpern, H., Spelbring, D.R. and Pelizzari, C.A. Beams's eye view volumetrics: An aid in rapid treatment plan development and evaluation. Int.J.Radiat.Oncol.Biol.Phys. 23:367-375, 1992.
  • [9] Bendl, R., Keller-Reichenbecher, M.A., Hoess, A., Schlegel,W. The Spherical View - A Tool for the Interactive Optimization of Irradiation Directions. Proceedings of the XIIth International Conference on the use of Computers in Radiation Therapy, this volume, 1997.
  • [10] Lyman, J.T. Complication probability as assessed from dose-volume histograms. Radiat. Res. 104 : 13- 19,1985
  • [11] Kutcher, G.J., and Burman, C. Calculation of complication probability factors for non-uniform normal tissue irradiation: the effective volume method. Int.J.Radiat.Oncol.Biol.Phys. 16: 1623-1630, 1989
  • [12] Webb, S., Nahum, A.E. A model for calculating tumor control probability in radiotherapy including the effects of inhomogenous distribtions of dose and clonogenic cell density. Phys.Med.Biol. 38 : 653- 666, 1993

to top