Clinical implementation of hyperthermia treatment planning guided steering: A cross over trial to assess its current contribution to treatment quality

Abstract

Purpose: To assess the current feasibility of online hyperthermia treatment planning guided steering (HGS) and its current contribution to treatment quality in deep hyperthermia for locally advanced cervical cancer.

Materials and methods: 36 patients were randomized to receive either their second and fourth (arm A) or their third and fifth (arm B) hyperthermia treatment of the series with the aid of HGS. The other treatments were conducted according to the Rotterdam Empirical Steering Guidelines (RESG).

Results: During period I (second and third treatment of the series) similar results were found for HGS and RESG with a slight, non-significant difference found in favour of HGS. The average temperature T50 was 40.3°C for both (p = 0.409) and the dose parameter CEM43T90 was 0.64 for RESG and 0.63 for HGS (p = 0.154). However, during period II (fourth and fifth treatment of the series) HGS performed less well, with significant lower thermal dose parameters, minimum, mean and maximum intraluminal temperatures, tolerance measures and net integrated power. T50 was 40.4°C after RESG and 40°C after HGS (p = 0.001) and CEM43T90 0.57 and 0.38 (p = 0.01) respectively.

Conclusion: We found that the procedure of online treatment planning guided steering is feasible. For maximal exploitation of its possibilities, however, better control and understanding of several patient, tumour and technical parameters is required. This study has been very helpful in identifying some of the challenges and flaws that warrant further investigation in the near future, such as patient positioning and the prevention of hotspot-related complaints.

• hyperthermia
• clinical trial
• treatment planning
• cervix cancer

Introduction

In the Netherlands, combined radiotherapy and deep hyperthermia has been part of regular healthcare for patients with locally advanced cervical cancer since 1996. Several randomised trials showed that the addition of deep hyperthermia to radiotherapy improves local control and survival for these patients [1–7] and most recently we demonstrated a 5-year local control rate of 53% [8]. Not withstanding this encouraging result in a group of patients with relative poor prognosis, i.e. large primary tumours, there is still ample room for further improvement of treatment outcome and we should continue to search for better treatment strategies [9], [10].

In this perspective, the finding of a significant correlation between the thermal dose delivered during treatment and patient outcome in a group of 420 patients with locally advanced cervix cancer treated with radiotherapy and deep hyperthermia clearly opens a window for further research [11]. This thermal dose-effect implies that better results should be obtained when higher thermal dose levels can be delivered. Obviously, the most elegant possibility to increase the thermal dose delivered is to aim for more tumour-selective and patient-specific heating than is currently achieved.

At present, most deep hyperthermia treatments are applied empirically, i.e. experience and dedication of the treatment team plays a major role in the final treatment quality. In general, the occurrence of hot spots, or areas of discomfort to the patient due to uncomfortable temperatures locally, limit temperatures achieved during hyperthermia.

Commonly, the strategy to manage hot spots is to apply a short break in the power applied, followed by adjustment of phase and amplitude settings to the antennas in order to steer the energy away from the hot spot. The precise approach of this strategy and thereby its effectiveness varies from centre to centre. Of course, a more objective approach would be preferable as it would allow a more systematic strategy and would also enable transfer of knowledge between centres and education of new staff. If the clinical application of such a systematic strategy were useful and effective, it would mean a major step forward. For the first time in the history of hyperthermia standardisation, improvement of treatment quality, a priori assessment of potential quality of treatment and treatment quality in centres new to the field can be expressed as an objective quality index. For these reasons, a hyperthermia treatment planning system is considered a great aid as hyperthermia treatment planning can help us better understand the effects of phase and amplitude adjustments on power and temperature distribution and even predict the effect of the adjustments during treatment [12–15]. Consensus exists that the combination of hyperthermia treatment planning with optimisation of treatment settings to maximise power deposition in the tumour and minimise hot spots will improve temperatures in deep hyperthermia [16].

The use of hyperthermia treatment planning is becoming common practice in hyperthermia [12–19], but for its online use during treatment, an optimisation routine is necessary that not only optimises power deposition in the tumour, but also reduces deposition in a complaint-related area (hyperthermia treatment planning guided steering, or HGS) [20]. Although not yet demonstrated in a clinical situation, the development of HGS for standardisation and improvement of treatment quality is a very important step in the further development of hyperthermia. Such a strategy would promote uniformity of treatment quality and comparison of treatments among the various institutes applying deep hyperthermia. On the other hand, the preparation process is time consuming and labour intensive and our current treatment approach (i.e. following the RESG) [14] proved its effectiveness in several clinical trials. Further, the RESG are based on decades of clinical experience, and it will be difficult to improve its results with a new technique that has never been used in a clinical setting before. As a first step we designed a clinical trial to compare the two treatment approaches in terms of temperatures achieved during treatment, thermal dose delivered and tolerance measures. This study provides us with an assessment of the current status and performance of HGS in a clinical situation and shows how clinical results compare to our gold standard defined by the RESG.

Materials and methods

Clinical background

For patients with locally advanced cervical cancer, five hyperthermia treatments of 90 min each are planned for each patient during the period of external beam radiotherapy. For all hyperthermia treatments in this study, the BSD2000-3D system (BSD Medical systems, Salt Lake City, UT) was used with the Sigma-60 applicator [21–23]. In Rotterdam, the standard operating frequency of the Sigma-60 is 77 MHz. The treatment is started at a power output of 400 W and was increased with steps of 100 W for every 5 mi as long as the patient has no hotspot-related complaints or normal tissue temperatures do not exceed 43°C. In case of hotspot-related complaints or normal tissue temperature > 43°C, the power is briefly turned off until the discomfort subsides or temperature is below 43°C, and phase, amplitude or frequency settings are adjusted to prevent recurrence. The further increase of power is not resumed until a complaint-free period of 5 min has been established. These principles were maintained over all treatments in this study. For a detailed description of the procedure and its rationale we refer to the paper of Van der Wal et al. [14].

Study design

For this study, all patients with locally advanced cervical cancer and an indication for deep hyperthermia were eligible if thermometry could be performed in the bladder, vagina and rectum. After informed consent, patients were randomised to receive either the second and fourth or the third and fifth hyperthermia treatment with HGS. A cross-over design was chosen because interpatient variation was observed as larger than the intrapatient variation as we learned from previous data [9]. Further, as the effect of hyperthermia treatment on intraluminal temperatures is short lived, it is unlikely that the outcome of a previous treatment influences the outcome of a consecutive treatment (i.e., probably no carry over effect). To account for the influence of progression of the treatment series on patient tolerance, both treatments were repeated per individual patient.

The first treatment was excluded from the study in order to allow the patient to become acquainted with the principles of the hyperthermia treatment and equipment.

Primary endpoints for this study were temperature, thermal dose and treatment-limiting hot spots. For temperature, we chose to use the T20 (the temperature exceeded by 20% of monitored sites per patient in the bladder, vagina and rectum), T50 (temperature exceeded by 50% of monitored sites per patient in the bladder, vagina and rectum) and T90 (temperature exceeded by 90% of monitored sites per patient in the bladder, vagina and rectum). For thermal dose, we chose CEM43T90 (cumulative equivalent minutes of T90 at 43°C as described by Fatehi et al. [22] and TRISE (a custom made thermal dose parameter based on T50 and the duration of heating). This second parameter has been shown retrospectively to correlate very well with treatment outcome in our patient group [11].

For treatment-limiting hot spots, we chose the number of off-switches, the total duration of off-switches and the time from start of treatment to first complaint as measures. An off-switch is defined as turning off the power of the BSD-2000 system longer than 20 seconds to reduce a hotspot-related complaint. Shorter off-switches are mostly caused by hyperthermia staff entering or leaving the treatment room and do not affect the intraluminal temperature profile during treatment.

Further, we chose the net integrated power as described by Fatehi et al. [24] as a secondary outcome measure, because an increase in net integrated power is expected to be accompanied by an increase in target temperature [16], [24].

Temperature and thermal dose data preparation

For thermometry, Bowman probes (BSD Medical Systems) were placed in 5 French polyethylene closed-tip catheters (William Cook Europe, Bjaeverskov, Denmark). These closed-tip catheters were placed in the patient's rectal, vaginal and bladder lumen. The rectal thermometer was inserted to a maximum insertion depth of 15 cm and the vaginal thermometer is inserted until the tip touches the cervix. The bladder thermometer is inserted alongside the urine catheter. The tips of both the urine catheter and the thermometer reach 4 cm inside the bladder; this insertion depth is fixed because the balloon is inflated after insertion to keep the urine catheter in place. The insertion depths are checked using a tape measure after insertion and reproduced during consecutive treatments of the same patient.

Thermal mapping was performed every 5 min with a step size of 1 cm and a maximum map length of 14 cm. Based on the temperatures measured intraluminally, several treatment parameters were calculated using the Rotterdam hyperthermia thermal modulator, which has been described elsewhere in detail [25].

Current treatment approach using the RESG

Preparation

Currently, all patients are positioned in the same way in the Sigma-60 applicator in the anterior-posterior and lateral directions. The preferred craniocaudal position is derived from the Computed Tomography (CT) scan, made for radiotherapy treatment planning. From this CT scan the distance from the centre of the tumour to a bony landmark, in this case the pubic bone, is calculated. The patient is positioned so that the centre of the pelvis is in the centre of the S60 applicator. In the craniocaudal direction, the patient is positioned such that the tumour centre is located 4 cm caudal to the centre of the Sigma-60 applicator. The start-up settings for phase and amplitude are the same for every patient, namely (0, 0) for phase and 100% amplitude for all BSD channels.

Optimisation during treatment

The RESG state that in case the patient has a hotspot-related complaint, the preferred order of steering actions is: phase steering (in 2-cm steps), amplitude steering (in 20% steps) and finally frequency steering (in 10 MHz steps). In addition, phase steering is thought to be more appropriate in case of pressure-like, deep-seated complaints and amplitude steering in case of burning, superficially located sensations. For example, when a patient complains of uncomfortable pressure on the abdomen, the phase is shifted 2 cm to the back and when a patient complains of a burning sensation to the buttocks, the amplitude of the dorsal BSD antenna is lowered by 20%.

The power is only lowered when phase, amplitude and frequency steering proved ineffective. Besides avoiding and diminishing hotspot-related complaints, we also aim for a homogeneous intraluminal temperature distribution during treatment by means of phase and amplitude steering [14].

Treatment approach using HTP guided steering (HGS)

Preparation

Prior to the first hyperthermia treatment, a CT scan was made of each patient lying in a hyperthermia treatment position, which was identical to the position in the Sigma-60 applicator, i.e. in a BSD sling system specially mounted on the CT scanner for the hyperthermia treatment planning CT scans. All CT scans were made using a multi-slice CT scanner (Siemens Somatom Sensation Open, Siemens Medical Solutions, Malvern, PA) with a slice distance of 0.5 cm. The scanned length of the patient had to be at least 80 cm to cover the length of the Sigma-60 with 10 cm extra at each end (cranial and caudal). The methods employed for hyperthermia treatment planning have been described elsewhere in detail [12], [13], [20], [26]. After resampling the CT data to 256 × 256 × 80 pixels, the following tissue types were segmented: tumour, muscle, fat, bone, liver, spleen, kidney, heart, lung, uterus, intestine, stomach, bubbles of air in the bowel system and vagina. Note that we segmented the actual anatomy instead of taking a single permittivity and conductivity as an average for the whole intestine in the pelvic region. The large vessels were not segmented separately but as muscle because of the lack of specific perfusion information and the fact that SAR and not temperature optimisation was performed. The permittive and conductive properties assigned to the specific tissue types were derived from Gabriel et al. [27] and are described in Table I. Segmentations were performed by an experienced physician (M.F.) who did all segmentations in order to promote uniformity.

Table I.  Dielectric parameters used for treatment planning.

Tissue	εr	σ (S/m)	ρ (kg/m^3)
Fat	13	0.07	900
Muscle	69	0.70	1000
Bone	16	0.06	1600
BSD sling support rods	1	0	1000
Target	69	0.70	1000
Bladder	24	0.29	1000
Heart	99	0.70	1000
Intestine	108	1.62	1000
Kidney	109	0.77	1000
Liver	75	0.46	1000
Lung	35	0.71	500
Spinal cord	6	0.04	1000
Spleen	101	0.77	1000
Stomach	82	0.89	1000
Uterus	69	0.70	1000
Vagina	69	0.70	1000

After construction of a tetrahedral model, the SAR distribution inside the patient was calculated using the finite element method (FEM)-module of Sigma HyperPlan (Dr. Sennewald Medizintechnik, Munich). Then this SAR distribution was optimised using a custom-made complaint-adaptive power density optimisation tool [20] providing us with patient-specific optimal treatment settings to start a treatment.

Patient positioning

For patient positioning during HGS treatments, the preferred craniocaudal position was derived from the CT scan made for hyperthermia treatment planning similar to the currently used method. The anterior-posterior distances of the patient's contour to the applicator were measured in the Sigma HyperPlan model and, as accurately as possible (preferably < 1 cm) [12], [17], [28] reproduced in the clinical setting using two ultrasound measurement probes integrated in the Sigma-60 ring. Before the first HGS treatment, an initial SAR optimisation was performed, providing us with patient-specific start-up settings for phases and amplitudes.

Optimisation during treatment

During the HGS treatments, hotspot-related complaints reported by the patient were dealt with similar to RESG. That is, power was turned off until the patient was comfortable again and then power was turned back on with the adjusted treatment settings. The difference lies in the adjusted treatment settings; the exact amplitude and phase settings were dictated by custom-made optimisation software [20] during HGS treatments and not by the RESG. We needed to define specific hotspot-related regions in the model to allow for the limitation of SAR in that specific region, while still optimising SAR in the tumour region. In case of a hotspot-related complaint in the abdomen, a constraint was assigned to the ventral abdominal muscles and new treatment settings were calculated with optimal power delivery to the tumour and minimal power to the ventral abdominal muscles. Homogeneity of measured intraluminal temperatures was not a goal during these treatments.

Statistical analysis

Prior to the start of the study a power analysis showed that 36 patients would be needed to show a 0.3°C difference in temperature measures with this double cross-over design with a power (1-β) of 80% and a significance level (α) of 95%.

First, we compared treatment parameters between the arms of the study using a T-test for two independent samples (comparison 1). This was done separately for period I (the second and third treatments) and II (the fourth and fifth treatments). The aim of this analysis was to assess whether a carry-over effect was present. In case no carry-over effect was present (i.e., no difference between the two randomisation arms), the data were analysed according to the cross-over design of the study [29]. If a carry-over effect was present, the data should be analysed according to a standard parallel group design, i.e. restricted to the first episode of period I and/or II.

According to the cross-over design, we compared the patient's first RESG treatment with the patient's first HGS treatment and the patient's second RESG treatment to the patient's second HGS treatment using a paired t-test, disregarding the arm of randomisation (comparison 2).

To test whether effect estimates differ between period I and II, a regression model was designed with treatment (RESG versus HGS), and period (I versus II) as covariates and an additional term for interaction between treatment and period (comparison 3).

For all statistical analyses, STATA version 10.1 was used (StataCorp, College Station, TX). P-values below 0.05 were considered significant. For comparison 3, the possible correlation between measurements from the same patient in the course of her treatment was taken into account by including a random effect for the intercept in the models. This was done by using a linear mixed regression model (STATA module xtmixed, StataCorp) [30].

Results

Patient and tumour characteristics of the 36 patients included in this study are summarised in Table II. No significant differences were observed between the two arms as assessed using a t-test.

Table II.  Patient and tumour characteristics.

		Arm A	Arm B
FIGO stage	IB2	1 (5%)	3 (19%)
	IIA	0 (0%)	1 (6%)
	IIB	6 (30%)	3 (19%)
	IIIA	2 (10%)	1 (6%)
	IIIB	6 (30%)	1 (6%)
	IVA	2 (10%)	3 (19%)
	IVB	3 (15%)	4 (25%)
WHO performance status	0	12 (65%)	10 (69%)
	1	7 (35%)	5 (18%)
	2	0 (0%)	2 (13%)
Nodal status	Nx	9 (45%)	4 (25%)
	N0	3 (15%)	7 (44%)
	N1	8 (40%)	5 (31%)
Age	Mean (range)	60 (30–84)	55 (35–79)

Arm A, second and fourth hyperthermia treatment of the series with the aid of hyperthermia treatment planning guided steering (HGS); Arm B, third and fifth hyperthermia treatment of the series with the aid of hyperthermia treatment planning guided steering, FIGO, International Federation of Gynaecology and Obstetrics, *FIGO IVB includes patients with para-aortic and/or supraclavicular metastasis; WHO, World Health Organisation; Nx, lymph node status unknown; N0, no pathological lymph nodes detected; N1, pathological lymph nodes detected.

One patient did not receive any HGS treatments because of a rapid deterioration of her clinical condition during treatment due to gastro-intestinal toxicity. In Table III, the model properties for each of the 35 remaining patients are summarised.

Table III.  Average model properties for all 35 patients who received HGS treatments.

Patient	Number of tetrahedra	Maximum edge length of a tetrahedra		Maximum edge length of a tetrahedra		Tetrahedra volume
		Mean (cm)	SD (cm)	Mean (cm)	SD (cm)	Mean (cm^3)	SD (cm^3)
1	2 19 262	3.48	1.97	1.81	1.14	4.13	10.18
2	2 34 484	3.63	2.15	1.92	1.25	5.01	12.00
3	2 12 570	3.80	2.14	2.01	1.24	5.31	12.00
4	1 97 587	3.47	1.86	1.82	1.08	3.86	9.47
5	2 00 939	3.51	1.98	1.84	1.15	4.22	10.19
6	1 98 775	3.56	2.00	1.87	1.16	4.37	10.50
7	2 01 501	3.45	1.92	1.80	1.12	3.94	9.69
8	1 82 699	3.47	1.82	1.81	1.06	3.71	8.83
9	2 06 770	3.36	1.83	1.75	1.07	3.58	9.04
10	2 23 556	3.33	1.88	1.75	1.08	3.65	9.36
11	2 07 982	3.26	1.71	1.69	1.00	3.13	8.05
12	2 00 855	3.54	2.00	1.87	1.16	4.36	10.37
13	1 98 907	3.30	1.76	1.72	1.03	3.30	8.33
14	2 01 024	3.44	1.87	1.80	1.10	3.85	9.57
15	2 07 030	3.48	1.96	1.82	1.15	4.15	10.07
16	2 33 736	3.15	1.88	1.64	1.09	3.31	8.79
17	2 40 469	3.25	1.97	1.69	1.15	3.73	9.84
18	3 53 697	2.52	1.65	1.29	0.93	1.98	6.74
19	2 83 914	2.70	1.73	1.38	1.00	2.33	7.17
20	2 50 641	2.89	1.69	1.47	0.99	2.47	7.10
21	2 04 519	3.29	1.80	1.72	1.05	3.38	8.58
22	2 96 232	2.65	1.78	1.37	1.02	2.35	7.43
23	2 78 368	2.95	1.89	1.50	1.10	3.00	8.71
24	2 66 723	2.85	1.77	1.44	1.03	2.58	7.53
25	2 75 863	2.88	1.81	1.46	1.06	2.72	7.97
26	2 59 504	2.98	1.83	1.52	1.08	2.94	8.29
27	2 68 183	2.94	1.80	1.49	1.06	2.80	8.12
28	2 68 499	2.84	1.80	1.44	1.06	2.66	7.81
29	3 01 996	2.87	1.88	1.45	1.10	2.90	8.73
30	3 66 561	2.58	1.75	1.35	0.99	2.30	7.67
31	2 73 816	3.11	1.92	1.58	1.12	3.33	9.35
32	3 02 384	2.64	1.77	1.36	1.01	2.34	7.39
33	2 69 534	2.94	1.76	1.49	1.03	2.71	7.84
34	2 88 728	2.88	1.72	1.46	1.01	2.55	7.66
35	3 66 685	2.62	1.63	1.37	0.92	2.11	7.11
Average over all patients:
	2 46 391	3.15	1.85	1.63	1.08	3.32	8.83

In Table IV, the various outcome measures of this study are reported by type of treatment (RESG or HGS) for periods I and II and for arm A and arm B separately.

Table IV.  Estimate (SD) for the outcome measures per arm of the study.

		Period I			Period II
Variable		RESG	HGS	total	RESG	HGS	total
Duration(min)	Arm A	88.2 (8.3)	89.5 (1.7)	88.8 (6.0)	88.4 (6.0)	90.0 (0.9)	89.0 (4.3)
	Arm B	86.6 (8.9)	89.7 (1.2)	88.1(6.5)	86.94 (7.8)	85.9 (13.0)	86.5 (10.5)
				∞0.6218			∞0.1626
	Total	87.3 (8.6)	89.5 (1.6)	88.5 (6.2)	87.7 (6.9)	88.1 (8.6)	87.9 (7.7)
			‡0.1445			‡0.7508
CEM43T 90 (min)	Arm A	0.61 (0.80)	0.73 (0.64)	0.67 (0.71)	0.55 (0.53)	0.42 (0.42)	0.48 (0.48)
	Arm B	0.69 (1.10)	0.55 (0.48)	0.62 (0.85)	0.64 (0.44)	0.33 (0.31)	0.49 (0.41)
				∞0.7853			∞0.9826
	Total	0.64 (0.94)	0.65 (0.57)	0.65 (0.77)	0.57 (0.48)	0.38 (0.37)	0.49 (0.45)
			‡0.9110			‡0.0124
TRISE (°C)	Arm A	3.33 (0.92)	3.40 (0.63)	3.37 (0.78)	3.22 (0.83)	2.97 (0.80)	3.10 (0.81)
	Arm B	2.98 (0.80)	3.22 (0.67)	3.10 (0.73)	3.34 (0.71)	2.78 (0.73)	3.08 (0.76)
				∞0.1493			∞0.9149
	Total	3.17 (0.89)	3.33 (0.64)	3.25 (0.77)	3.26 (0.77)	2.89 (0.76)	3.09 (0.79)
			‡0.1544			‡0.0044
T20 (°C)	Arm A	40.9 (0.8)	40.9 (0.7)	40.9 (0.7)	40.9 (0.8)	40.6 (0.8)	40.7 (0.8)
	Arm B	40.7 (0.9)	40.8 (0.7)	40.7 (0.8)	41.0 (0.8)	40.4 (0.7)	40.7 (0.8)
				∞0.2741			∞0.9385
	Total	40.8 (0.9)	40.9 (0.7)	40.8 (0.8)	40.9 (0.8)	40.5 (0.8)	40.7 (0.8)
			‡0.6452			‡0.0006
T50 (°C)	Arm A	40.4 (0.8)	40.4 (0.6)	40.4 (0.7)	40.3 (0.8)	40.0 (0.8)	40.1 (0.8)
	Arm B	40.1 (0.9)	40.2 (0.7)	40.2 (0.8)	40.5 (0.7)	39.9 (0.6)	40.2 (0.7)
				∞0.2499			∞0.6547
	Total	40.3 (0.9)	40.3 (0.7)	40.3 (0.8)	40.4 (0.8)	40.0 (0.7)	40.2 (0.8)
			‡0.4093			‡0.0009
T90 (°C)	Arm A	39.5 (0.8)	39.7 (0.6)	39.6 (0.7)	39.4 (0.8)	39.2 (0.8)	39.3 (0.8)
	Arm B	39.4 (0.8)	39.5 (0.7)	39.4 (0.8)	39.7 (0.6)	39.1 (0.6)	39.4 (0.6)
				∞0.4203			∞0.5859
	Total	39.4 (0.8)	39.6 (0.6)	39.5 (0.7)	39.5 (0.7)	39.2 (0.7)	39.4 (0.7)
			‡0.1033			‡0.0051
NOS	Arm A	13.2 (5.2)	15.3 (6.9)	14.2 (6.1)	13.8 (4.9)	16.1 (6.1)	14.9 (5.6)
	Arm B	14.0 (5.5)	14.3 (6.4)	14.1 (5.9)	14.2 (6.7)	18.6 (6.8)	16.3 (7.0)
				∞0.9469			∞0.3528
	Total	13.6 (5.3)	14.9 (6.6)	14.2 (6.0)	14.1 (5.7)	17.1 (6.4)	15.5 (6.2)
			‡0.2754			‡0.0245
DOS (min)	Arm A	8.4 (3.7)	10.5 (4.9)	9.4 (4.4)	8.9 (3.1)	12.6 (6.4)	10.8 (5.3)
	Arm B	8.8 (3.9)	10.8 (5.8)	9.7 (5.0)	7.5 (4.1)	12.6 (4.8)	10.0 (5.1)
				∞0.7770			∞0.5328
	Total	8.5 (3.8)	10.6 (5.3)	9.6 (4.6)	8.3 (3.6)	12.6 (5.7)	10.4 (5.2)
			‡0.0300			‡0.0002
TTFC (min)	Arm A	26.4 (11.2)	28.8 (11.1)	27.6 (11.1)	24.9 (10.3)	28.2 (14.7)	26.6 (12.6)
	Arm B	25.6 (13.3)	22.6 (8.9)	24.2 (11.3)	22.5 (9.7)	20.5 (6.2)	21.5 (8.1)
				∞0.2106			∞0.0591
	Total	26.0 (12.2)	26.1 (10.5)	26.0 (11.2)	23.8 (10.1)	24.8 (12.2)	24.3 (11.1)
			‡0.9890			‡0.6455
NIP (kJ)	Arm A	3028 (651)	2972 (504)	3000 (575)	3051 (486)	2776 (480)	2913 (497)
	Arm B	2890 (692)	3009 (601)	2948 (642)	2919 (548)	2602 (695)	2766 (633)
				∞0.7166			∞0.2741
	Total	2941 (655)	2988 (539)	2977 (601)	2987 (517)	2702 (579)	2849 (561)
			‡0.6080			‡0.0002

‡p-value for comparison of RESG versus HGS, disregarding the arm of the study.

∞p-value for comparison of arm A versus arm B, disregarding the number of the treatment.

Comparison 1: Cross-over effect

In Table IV, p-values in the 5th and 8th column of Table IV (both marked with ∞) represent the significance levels for the difference between arm A and arm B. So we can conclude that no carry-over effect is present in this study as the differences in treatment outcome between arm A and arm B are insignificant.

Comparison 2: RESG versus HGS effect

In Table IV the p-values in the 4th and 7th column (both marked with ‡) represent the significance levels for the difference in treatment outcome measures between RESG and HGS, irrespective of whether the treatments were performed in arm A or arm B.

In period I, only the duration of off-switches is significantly longer in the HGS treatments with a difference of 2.1 min (p = 0.03), indicating less efficient coping with hotspot-related complaints during HGS treatments. Further, there were favourable trends towards a longer duration of treatment (87.3 min for RESG and 89.5 min for HGS, p = 0.14), higher net integrated power (2941 kJ for RESG and 2988 kJ for HGS, p = 0.61), higher TRISE (3.17°C for RESG and 3.33°C for HGS, p = 0.15) and higher CEM43T90 (0.64 min for RESG and 0.65 min for HGS, p = 0.91) during HGS-treatments, although these trends were not significant (Table IV).

The analysis for period II shows a different picture. The HGS treatments in this period show significantly lower thermal dose. The average CEM43T90 was 0.57 min for the RESG treatments in period II and 0.38 for the HGS treatments in that period (p = 0.01, Table IV). For the average TRISE, a similar significant difference was found; 3.26°C for RESG treatments and 2.89°C for HGS treatments in period II (p = 0.00, Table IV). Further, intraluminal temperatures were significantly lower in period II (T20 with a 0.4°C difference, T50 with 0.4°C and T90 with 0.3°C, Table IV). Figure 1 illustrates the differences in thermal dose and temperature measures per period and per treatment type. It becomes clear that although the differences are statistically significant, their clinical relevance may be minimal. Figure 2 illustrates the variation in T50 per period and per patient, and it shows that there is considerable variation between patients; some do worse with HGS compared to RESG and some do better.

Figure 1. Outcome per period and per treatment type (with 95 % confidence intervals). (A) Temperature measures; (B) Thermal dose parameters thermal dose parameters. Period I, first part of hyperthermia treatment series, i.e. treatments 2 and 3; Period II, second part of hyperthermia treament series, i.e. treatments 4 and 5; RESG1, currently used treatment approach following the Rotterdam Empirical Steering Guidelines ¹⁴ during period I; HGS1, treatment approach using hyperthermia treatment planning guided steering during period II; RESG2, currently used treatment approach following the Rotterdam Empirical Steering Guidelines [14] during period I; HGS2, treatment approach using Hyperthermia treatment planning Guided Steering during period II; CEM43T90, cumulative equivalent minutes of T90 at 43°C in minutes as described by Fatehi et al. [24]; TRISE, a local custom-made thermal dose parameter based on T50 and the duration of heating¹¹ in °C; T20, temperature exceeded by 20% of the monitored sites in bladder, vagina and rectum; T50, temperature exceeded by 50% of the monitored sites in bladder, vagina and rectum; T90, temperature exceeded by 90% of the monitored sites in bladder, vagina and rectum.

Figure 2. T50 per period and per treatment type, interpatient variation. RESG1, currently used treatment approach following the Rotterdam Empirical Steering Guidelines [14] during period I; HGS1, treatment approach using hyperthermia treatment planning guided steering during period II; RESG2, currently used treatment approach following the Rotterdam Empirical Steering Guidelines [14] during period I; HGS2, treatment approach using hyperthermia treatment planning guided steering during period II; T50; temperature exceeded by 50% of the monitored sites in bladder, vagina and rectum.

Hot spot complaints seem less well handled in period II (number of off-switches was increased by 3, duration of off-switches was prolonged with 4, 3 minutes, p = 0.02 and 0.00 respectively) and net integrated power decreased (279 kJ more was administered during RESG treatments, p = 0.00).

Comparison 3: Differences between treatment period and type of treatment

The interaction between treatment period and type of treatment is significant for TRISE (p = 0.001), T20 (p = 0.002), T50 (p = 0.001) and T90 (p = 0.001), indicating a significant difference in the effect of HGS versus RESG between period I and period II. This is in accordance with the results of comparison 2, where the effects of HGS and RESG during period I and period II are analysed separately.

Learning effects encountered during study

Advanced understanding of applying hyperthermia treatment planning optimization

During the study it became clear that our primary optimisation method (Opt1 from Canters et al.) [20] insufficiently dealt with hotspot-related complaints to allow for a meaningful and swift reaction to clinical situations. We therefore adjusted the optimisation method to not only optimise power deposition in the tumour, but also to minimise power deposition in a specific hotspot-related area in the model while maximising power deposition in the tumour (Opt2 from Canters et al.) [20]. The main difference between the two methods is that Opt1 considers only SAR in the tumour region, while Opt2 also takes hot spots into account. In phantom studies, Opt2 showed better hot spot reduction and spatial control, and based on this finding we switched from Opt1 to Opt2 in this study. As a result, the first five patients who entered the study were treated using Opt1 during the HGS treatments. The other 30 were treated using Opt2.

Improved patient positioning

Another problem we encountered during the course of the study was that the accuracy of currently used positioning techniques was somehow inadequate for use in conjunction with a HGS. When trying to reproduce the patient's position from the CT-based computer model to the actual patient position in the Sigma-60 applicator, we encountered problems with patients’ legs touching the outer rim of the Sigma-60 when the anterior-posterior position measured in the model was copied to clinical situation. A closer look at our current patient positioning protocol in clinical practice and the protocol used for hyperthermia treatment planning CT scans, revealed that most patients were positioned much more cranially during the CT scan than during treatment. The reason for this was our demand for a single continuous CT scan for the hyperthermia treatment plan, which was only possible when the patient was positioned more cranially in the CT scanner. This problem of patient position alignment was solved when specific attention was paid to the craniocaudal positioning of the patient in the BSD hammock, so no more problems were encountered with patient positioning.

Outcome for patients who were correctly positioned

When repeating comparison 2 for patients who were correctly positioned, no differences in outcome measures were observed when comparing them to the results of comparison 2 for the whole group of patients. For period I only, the duration of off-switches is significantly longer for HGS treatments (p = 0.03), all other differences were not significant. For period II, again HGS tends to lead to more and longer off-switches, lower thermal dose and lower temperatures compared to RESG. The same outcome we observed in the whole group, namely that results are similar for HGS and RESG for period I, but during period II HGS performs less well, also holds true for this subgroup.

Discussion

In this article we present our first experience with taking hyperthermia treatment planning guided steering, or HGS, to the clinic. HGS proved to be feasible in every day clinical practice. Early on in a treatment series, HGS performs as well as RESG and in view of the fact that the RESG were developed based on years of clinical experience, this is a very worthwhile result.

During the first part of a treatment series (period I, second and third treatment) only the duration of off-switches, a measure for treatment-limiting hot spots, was significantly longer during HGS treatments. During each hyperthermia treatment, the power is turned off when a patient shows signs or symptoms indicating a hotspot-related complaint. During RESG treatments, the power is turned on again when the patient indicates the complaint has subsided. During HGS treatments, the power was turned on again when the complaint has subsided and new treatment settings were calculated with a custom-made add-on [20] to Sigma HyperPlan. The calculation time required by Sigma HyperPlan (up to 2 min) could well explain the difference in the duration of off-switches. For thermal dose parameters, maximum temperature and time to first complaint, a slight, non-significant difference in favour of HGS could be found for period I.

The analysis for period II (fourth and fifth treatment of the series) shows a more complicated picture. HGS treatments now show significantly lower power, intraluminal temperatures (with a difference of 0.4°C for T20, 0.4°C for T50 and 0.3°C for T90) and thermal dose (with a difference in TRISE of 0.37°C and in CEM43T90 of 0.19 min). Whether these differences have a clinical meaning, remains to be seen (Figure 1). Our previous thermal dose analysis showed a significant correlation between thermal dose parameters and treatment outcome, but with great dispersion of the data [11]. Further, it remains questionable whether intraluminal temperatures represent intratumoural temperatures as well in more tumour-selective heating (HGS) as in the more empirical regional heating that is obtained using the RESG. We must realise that changing heating strategy may cause historical correlations to no longer be valid. For example, Fatehi et al. showed good correlation between intraluminal and intratumoural temperatures, i.e. when treatment settings are adjusted to obtain a homogeneous intraluminal temperature distribution [31]. During HGS treatments, treatment settings are not adjusted to aim for a homogeneous intraluminal temperature distribution, but to obtain maximum SAR in the tumour. If this is done sufficiently selectively, this could paradoxically cause a decrease of intraluminal temperatures as a consequence of the more targeted treatment strategy.

In our previous thermal dose analysis of 420 patients it was already apparent that patients become harder to heat as treatment progresses (Figure 3) using the RESG [11]. A possible explanation for this finding is that as treatment progresses patient tolerance decreases due to the cumulating fractions of radiotherapy administered; acute radiation-induced toxicity and fatigue set in. Also, the applications of brachytherapy are usually administered in the fourth and fifth week of treatment, greatly increasing the sensitivity and tenderness of a patient's pelvic area. As this previously found difficulty with heating a patient as treatment progresses, which is also expected to play a role in this study, we decided to introduce the analysis per period (period I and period II) in order to account for this. The difference between RESG and HGS becomes much more apparent in period II, which could be explained by the fact that RESG is a much simpler optimisation model compared to HGS. RESG leaves more room for individual interpretation, making it more flexible and better equipped to deal with decreasing patient tolerance. Another important aspect which may affect the proper assessment of the performance of the HGS treatments in period II is whether the anatomical information as obtained from the pre-treatment CT scan is still valid. During the course of treatment, the tumour shrinks, patients may lose weight, and the chemical balance in the intestine may change due to diarrhoea. In parallel with the anatomical changes, the biological and physiological characteristics of the tumour will also change during treatment. All of these factors may affect the energy and temperature distribution in the patient and are not taken into account in the treatment planning at the start of treatment. These factors may very well explain why HGS performs less well as the treatment progresses and the original anatomy gets distorted.

Figure 3. Evolution of temperatures (A) and thermal dose (B) over treatment series based on the data of reference 11. (A) temperatures (°C); (B) Thermal dose (CEM43T90 in min, TRISE in °C). CEM43T90, cumulative equivalent minutes of T90 at 43°C as described by Fatehi et al. [24]; TRISE, a local custom-made thermal dose parameter based on T50 and the duration of heating [11]; T20, temperature exceeded by 20% of the monitored sites in bladder, vagina and rectum; T50, temperature exceeded by 50% of the monitored sites in bladder, vagina and rectum; T90, temperature exceeded by 90% of the monitored sites in bladder, vagina and rectum.

From a technological point of view HGS performs much more like an open-loop feedback system than RESG. As previously explained, HGS is critically dependent on the robustness of the input parameters at the start of a treatment. In contrast, the clinician's observation of the patient condition provides the clinician with an update of the subjective (although difficult to quantify) input parameters for RESG at the beginning of each successive hyperthermia treatment. Clearly, the design of future studies should include updating of the treatment planning based upon the changing anatomy, and if feasible, input of the changing biological and physiological characteristics of the tumour.

Although the results of this trial show that HGS in its current status can be of merit when applying deep hyperthermia, the 0.3°C improvement HGS this study was designed to detect could not be found. Since the study closed, we performed a number of theoretical studies that showed that with optimisation using the Sigma-60, the maximum SAR improvement that can be reached is within the order of 5%. Using the bioheat equation [32], this 5% SAR should lead to a rise in temperature of 0.2°C, an increase that is within the resolution of our currently used thermometry [27]. In retrospect, our estimated 0.3°C improvement using HGS may have been too high a goal with the hyperthermia equipment that we used.

Lessons learned from the clinical implementation of HTP guided steering

As to be expected when putting any new technique to clinical use, we encountered a number of challenges. Early on, we noticed that our first optimisation routine insufficiently addressed hotspot-related complaints reported by the patients. This prompted the development of a new optimisation routine that maximised power deposition in the tumour and minimised power in a specific hot spot related area [18].

We also encountered problems in patient positioning, which we were able to overcome with the currently available positioning techniques, although we would like to stress the importance of further improvements needed in this area. When correct patient positioning fails, high resolution optimisation procedures are useless.

A much mentioned drawback of hyperthermia treatment planning in general is the time-consuming nature of the process. In this study, one of the rules was that the CT scan made for hyperthermia treatment planning had to be made at least 3 days before the first study treatment took place. As we gained more experience with the segmentation process we were able to improve speed. The time required for segmentation was reduced from 8–9 hours per CT scan in the beginning to 3–4 hours near the end of the study. This may be further improved in the future using atlas-based segmentation. On average, calculation time was 15 hours for preparation and E-field simulation, a value which may change in time depending on computer speed and segmentation resolution.

Technical limitations

This study was designed to evaluate the efficacy of currently available hyperthermia treatment planning possibilities in the Sigma-60 applicator, with its inherent limitations. From the study by Canters et al. [28] the potential of HGS to optimise the SAR distribution in the Sigma-60 appears to be limited, due to the small number of degrees of freedom. The potential appearing from this model study could easily be lost due to inaccuracies in the hyperthermia treatment planning software, the dielectric constants and in the translation from model to clinic.

Two important limitations of the system we used in this study are the lack of optimal steering possibilities and the unknown influence of transforming networks. Also, the focus that is created by the BSD 2000 system and the Sigma-60 applicator is quite large, and with extreme settings, its performance decreases.

Clinical implications

We have no doubt that hyperthermia treatment planning is a necessary and inevitable next step in the development of hyperthermia as an oncological treatment modality. It enables patient-specific optimisation of treatment, which should eventually lead to a more standardised application of hyperthermia and better treatment quality. For now, we recommend the use of HGS for clinicians with no or limited experience in the field of hyperthermia, as this study shows that with the use of HGS clinical results can be obtained that are approaching our results with 18 years of experience.

Hyperthermia treatment planning is also a helpful tool in the evaluation of clinical indications; it may help clinicians decide in advance whether a tumour at a specific location can be heated to therapeutic temperatures or not. Further, it will be a great aid in education and training of new hyperthermia staff.

Hyperthermia treatment planning also proved to be a helpful tool in the development of new hyperthermia systems [33–37]. When a hyperthermia treatment planning system is used to develop a new system, the technical capabilities can be made better in line with the clinical demands.

Last but not least, hyperthermia treatment planning can be an important tool in more controlled treatment quality.

Future directions

We found that the procedure of online HGS is feasible. For maximal exploitation of its possibilities, however, better control and understanding of several patient, tumour and technical parameters is required.

As a whole, this trial has been very useful in terms of assessing what we currently can and cannot do with treatment planning for deep hyperthermia. Some lessons were quickly learned, while more time is needed for others. For example, it is mandatory to get more insight into relation between intraluminal temperatures with intratumoural temperature. One could argue that better focusing of energy in the target area could lead to a decline in intraluminal temperatures for some patients, and an increase in others, depending on patient anatomy and tumour vasculature and shrinkage.

Another point that requires further investigation is the relationship between a patient's hotspot-related complaint and a hot spot in the Sigma HyperPlan model, as temperature causes hot spots and not SAR, on which we optimised. This could, in part, explain our difficulties in clearing hotspot-related complaints during the HGS treatments. In addition, the indication of hotspot-related complaints by a patient is subjective by definition and in our experience there is great variation in how well patients are able to describe sensations in their body during hyperthermia.

Conclusion

In spite of the problems we encountered during this study and the inherent limitations due to the equipment and the current state of hyperthermia treatment planning, HGS performed equally well in treatments two and three when compared to the RESG based on our two decades of clinical experience. This study has been very helpful in identifying some of the challenges and flaws that warrant further investigation in the near future, such as patient positioning and the prevention of hotspot-related complaints. With the progress that has been made during this study, we hope to perfect the principle of hyperthermia treatment planning guided steering in the near future.

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