Hyperthermia classic commentary: ‘Simulation studies promote technological development of radiofrequency phased array hyperthermia’ by Peter Wust et al., International Journal of Hyperthermia 1996;12:477–494

Abstract

Simulation studies can estimate the potentials of regional hyperthermia based on the multi-antenna principle. They postulate an optimal parameter setup (phases, amplitudes) that should increase temperatures in heat treatments by 1 °C or more. However in praxi, slight inaccuracies and uncertainties accumulate in a way that the optimal adjustment is typically missed during a real treatment. The reasons for these errors are electrical and geometrical. A way out is an on-line control, which adapts the plans to the actual distributions. Magnetic resonance thermography is employed for on-line control, and control software has been developed. However, there is still a long way to reach the technological endpoint of multi-antenna systems for clinical applications.

• treatment planning
• online control
• annular-phased-array technique

The numerical studies in the referenced article [1] predicted that an increase in the number of antennas (channels) in radiofrequency applicators improves the temperature distributions in target regions selected for regional hyperthermia in the pelvis. This was an important argument to pursue the multi-applicator approach. However, since that time (now more than 12 years ago) we must concede that the progress to achieve higher temperatures reliably and safely in the heated tumours is only marginal, if any.

Since 1988 the SIGMA-60 applicator has been widely clinically applied and the limitations were evident. The study predicted that upgrading from a two-dimensional applicator (SIGMA-60) to a three-dimensional applicator with three rings (SIGMA-Eye) results in a significant temperature increase in the tumour region depending on the particular anatomical conditions. The improved patterns are predominantly utilising the 3D-configuration of a higher number of antennas enabling suppression of certain hot spots. We emphasise that higher temperatures in the tumour are mainly caused by increasing the total power.

Therefore, in the model the improved temperature distributions are not generated by adapting the power deposition patterns and keeping the total power constant, but instead by increasing the total power and simultaneously avoiding the hot spots occurring in typical locations during elevation of power.

In consequence, this approach needs a more sophisticated pattern monitoring and control not available at that time, and even not until today. For an easy execution of regional hyperthermia employing moderate (not too aggressive) power levels the standard patterns (e.g. in or near the resonance mode) might be more adequate and even more successful. These patterns have been described and characterised by planning studies [2].

The capabilities to intensify radiofrequency hyperthermia were confirmed and further clarified by supplementing simulation studies [3]. It was demonstrated that the most obvious leap was obtained by the transition from one antenna ring to two rings. Further improvements are generally associated with an increased number of antennas (up to 12 antennas per ring with single phase control) and higher frequencies (up to 200 MHz). However, it is also evident that the requirements on accuracy in positioning and pattern control are rising in parallel. Therefore, practical performance of heating is more demanding for more antennas, higher frequencies and necessarily higher power levels in order to finally achieve a higher temperature in the tumour.

Another observation in the modelling studies is also important: objective functions derived from SAR (specific absorption rate) distributions are not clinically useful, because they tend to generate hot spots. More successful are objective functions optimising the temperature distribution, e.g. calculated by solving the bio-heat transfer equation. This approach adds other uncertainties to the problem such as perfusions in the different tissues. In particular, the optimal adjustments for these solutions are difficult to predict by simple considerations.

Both complexity of the heating process and inability to find the optimum resulted in severe difficulties to transform these (theoretical) potentials in advance in real heat treatments. Therefore, after the clinical introduction of the SIGMA-Eye applicator some disappointment occurs, because no impressing temperature improvements could be documented.

Occasionally optimal phases derived from a planning procedure and adjusted for the clinical treatment even proved as counterproductive. Because they did not meet the real optimum they suffered from unexpected hot spots and/or restricted efficiency. Consequently, in practice, standard adjustments with no or only slight phase shifts were still preferred.

Now the question arises, why did we fail to realise our planning results until now? One (probably the most important) reason can be derived exactly from the simulation studies [3]. Inaccuracies of ±1 cm at 100 MHz result in considerable drops of the intratumoural temperatures T90, which are in the range of 0.5–1°C at 100 MHz and go up to 2–3°C at 200 MHz. These geometrical deviations correspond to phase errors of ±10° at 100 MHz and ±20° at 200 MHz. In other words, nearly all optimisation advantages are lost under practical conditions assuming errors in the cm range. The sensitivity increases with the frequency. In summary, uncertainties in phases, position and anatomy, and antenna characteristics are major hindrances to realising an optimum predicted by hyperthermia planning.

These error sources have been investigated by our group in recent years and documented by findings such as distortion of antenna functions [4] and strong coupling effects between antennas and feeding networks [5], [6]. We concluded that the phases and amplitudes in the feeding points of the antennas typically differ from the adjusted (i.e. optimised) parameters–and this is one important source error. To some extent, these deviations can be compensated by a re-adjustment of phases and amplitudes [4], [7]. However, a relevant difference between planned and actual SAR distributions remains even after correction of the channel adjustments (phases, amplitudes).

One particular difficulty associated with the SIGMA family of RF applicators is the use of antenna pairs, which (theoretically) leads to different phases (and voltages) in the feeding point of every single antenna. Inclusion of the complete feeding network into the planning process might improve the agreement between calculated and actual phases [8]. However, this is a very difficult, if not impossible task requiring an immense numerical effort in model formulation (antennas, networks) as well as software development. Until now we could not solve the problem to keep a SIGMA applicator under control with numerical approaches alone.

Note the fundamental difference to radiotherapy planning where we expect differences between planning calculations and actual dose distributions of less than 5%, and therefore could introduce highly sophisticated application techniques such as IMRT (intensity modulated radiotherapy) or dynamic arc therapy and others.

As a second attempt to overcome these uncertainties we implemented an E-field sensor in the water bolus to measure the electric field in phase and amplitude on a closed curve [9] and to derive the actual SAR distribution directly from these data. Also this method proved just as impractical under clinical conditions.

In a third approach we attempted to develop a multi-antenna applicator, which is by construction easier to model, and consequently an agreement between calculation and actual distribution is enforced, as investigated e.g. by Nadobny et al. [10]. Such an applicator would be advantageous and easier to handle, but a realistic concept competing with the SIGMA applicators has not been developed until now.

Is there any realistic approach to control multi-antenna applicators? We must understand that the registration of complete (i.e. three-dimensional) SAR- or temperature distributions is an essential precondition to compare plans and measurements and to find matching conditions.

Therefore, implementation of non-invasive MR thermography was a great success. For the first time we acquired entire 3D datasets in standard phantoms [11] and could directly compare calculated and measured SAR distributions derived from the MR temperature increases in every voxel [12].

Agreement between pre-planned and measured temperature patterns can be improved if phases, amplitudes and phantom position are corrected. However, there remained still a considerable gap under specific conditions. The largest differences between model and measurement are observed for symmetrical cases (e.g. a cylindrical phantom). Clearly, these differences between plans and measurements are dependent on the applicator. MR-compatible applicators might be more susceptible to such deviations because standard symmetrising network components based on ferrite cannot be used in an MR scanner.

We developed a numerical procedure to improve the agreement between theoretical and experimental data [13]. This is accomplished by generation of a complete control matrix V, which incorporates not only phases and amplitudes of antennas, but also antenna profiles. The latter can vary from antenna to antenna.

Starting with the best available model V₀ and any driving vector u (phases, amplitudes) we achieve , which typically is not properly describing our measurement. The control matrix is refined iteratively stepwise with every additional measurement V₁, V₂, etc., but the largest improvement arises with the first measurement (from V₀ to V₁). In consequence, we achieved acceptable agreement between actual measurements and planned patterns in homogeneous as well as heterogeneous phantoms. Here, we employed the equivalence of temperature increase from a defined reference state (i.e. MR-temperature) and SAR.

Using the corrected matrix V₁ (or V₂, V₃ …) we are now able to optimise the by adjusting phases (and amplitudes) with respect to any given objective function (e.g. maximising the ratio of power in the target relative to the power in the whole volume). We could demonstrate this optimisation process successfully in phantoms for several target locations.

We conclude that optimisation of antenna parameters works if the ‘correct model’ V₁ (V₂, V₃, …) is applied. Unfortunately, the improved model description V₁ derived from the original model V₀ is only local and depends on many parameters. It must be identified for every treatment situation (e.g. from day to day). On the other hand, we must start with a reasonable model V₀ to be successful. Other numerical approaches for MR-controlled hyperthermia were described by Stakhursky [14].

The key question is can we transfer this procedure to real patients? Non-invasive thermometry has been implemented successfully for certain tumour classes (soft tissue sarcomas, pelvic tumours) under clinical conditions [7], [15]. For these patients MR temperature datasets are online achieved and the method working in phantoms has been validated.

In case of the in vivo situation, the MR temperatures are influenced by more variables, among them at least perfusion and thermal and electrical tissue behaviour, and are therefore only partially determined by the SAR distribution. As a rule of thumb, MR temperature describes the temperature increase ΔT relative to an initial (reference) state (typically before switching on the amplifier system) and is given in first approximation by ΔT≅ const × SAR/perfusion in every tissue.

Interestingly, employing this feed-back control loop for in vivo MR temperature distributions, i.e. in patients, provides some improvements as well. We should, however, specify the criteria for such an improvement and differentiate the strategies:

1. Utilising the method available as feedback control, which basically regulates the SAR distribution, we attempt to concentrate more power into the target. This leads to a higher increment of the MR-temperature (relative to time and total power P) indicating an improved relative SAR distribution. By applying this method clinically, impressive increases of SAR/P have been already achieved in some patients (to be published). On the other hand, occasionally treatment-limiting hot spots have been clinically observed, which are known to be associated with SAR-based optimisations.

2. A more elaborated strategy, which basically follows the simulation procedures of Wust et al. and Seebass et al. [1], [3] is presently not implemented, but under development. Here the total power is gradually increased, which consequently enhances SAR and temperature in the tumour. Power-limiting hot spots must be circumvented by using this strategy. However, to predict and avoid hot spots successfully the full thermodynamic problem described by the bio-heat transfer equation (BHTE) has to be considered. Obviously, this is much more challenging, because the SAR = u^HV^HVu has to be adapted (via the system matrix V) and optimised (via the antenna vector u), but in addition the restriction of the BHTE has to be held. Further variables such as tissue-dependent perfusions and (probably less important) thermal diffusion play a role.

3. Every patient reacts individually to the power exposure. If discomfort occurs in a specific anatomical location, the power density must be further reduced. This is also possible with such an online procedure by input of suitable weighting factors.

4. Any online method to measure a treatment-related distribution might be useful, either from MRI or other imaging modalities. The algorithms developed can be generalised to every kind of dataset.

In summary, today, 12 years later, the conclusions of the theoretical studies are not refuted, but we have still not reached our goal. Assuming the capability to characterize and control temperature pattern use of higher frequencies and dedicated patterns might enhance the temperatures in the tumours considerably (1°C or more appears possible). It will cost much effort until the requirements for online control in clinical practice and (at best) for all indications are fulfilled. Scientific work of one institute will not be sufficient, and an integrated action is needed to be successful.

Acknowledgements

This work has been for many years supported by the Deutsche Forschungsgemeinschaft (DFG) in the Collaborative Research Project SFB 273 (1994-2002) and following projects (in particular WU 235/1-3).

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.