MRI-controlled transurethral ultrasound therapy for localised prostate cancer

Abstract

Minimally invasive treatments for localised prostate cancer are being developed with the aim of achieving effective disease control with low morbidity. High-temperature thermal therapy aimed at producing irreversible thermal coagulation of the prostate gland is attractive because of the rapid onset of thermal injury, and the immediate visualisation of tissue response using medical imaging. High-intensity ultrasound therapy has been shown to be an effective means of achieving thermal coagulation of prostate tissue using minimally invasive devices inserted into the rectum, urethra, or directly into the gland itself. The focus of this review is to describe the work done in our group on the development of MRI-controlled transurethral ultrasound therapy. This technology utilises high intensity ultrasound energy delivered from a transurethral device to achieve thermal coagulation of prostate tissue. Control over the spatial pattern of thermal damage is achieved through closed-loop temperature feedback using quantitative MR thermometry during treatment. The technology, temperature feedback algorithms, and results from numerical modelling, along with experimental results obtained in animal and human studies are described. Our experience suggests that this form of treatment is technically feasible, and compatible with existing MR imaging systems. Temperature feedback control algorithms using MR thermometry can achieve spatial treatment accuracy of a few millimetres in vivo. Patient-specific simulations predict that surrounding tissues can be spared from thermal damage if appropriate measures are taken into account during treatment planning. Recent human experience has been encouraging and motivates further evaluation of this technology as a potential treatment for localised prostate cancer.

• Feedback control
• high intensity focused ultrasound
• MRI thermometry
• prostate cancer

Introduction

High-intensity ultrasound therapy using externally focused ultrasound transducers has the potential to generate precise spatial patterns of thermal damage non-invasively in deep-seated tissue. Soft-tissue targets located behind bony structures or gas are, however, difficult to treat with externally focused transducers, necessitating approaches to compensate for increased absorption, reflection and distortion of the ultrasound beam [1–4]. An alternative strategy is to deliver high intensity ultrasound from devices inserted in body cavities or directly into soft tissue [5], [6]. These minimally invasive devices offer several advantages, such as the ability to use higher ultrasound frequencies due to reduced distance to the target and reduced sensitivity to motion because the devices are often in contact with the target region.

Spatial control over energy deposition from minimally invasive ultrasound heating applicators has been demonstrated through a number of transducer designs. The use of a linear array of transducers along the length of the device enables control over the axial pattern of energy deposition through adjustment of output power along the device [5], [7]. Angular control can be achieved by using directional transducers with device rotation [8–11], or arrays of elements arranged around the device to eliminate the need for rotation [12–14]. Control over radial energy deposition can be enhanced through the use of frequency modulation from individual elements [15]. These studies have demonstrated that many possible transducer configurations are capable of achieving control over spatial energy deposition around a transurethral device; however, none of these configurations has been evaluated clinically in human prostate.

The prostate gland is an organ for which intracavitary devices have been demonstrated effective for therapeutic ultrasound. Transrectal [16–18], interstitial [19–23] and transurethral [10], [13], [24–27] heating applicators have been designed to heat the prostate selectively while avoiding thermal damage in surrounding tissues. Transrectal devices employing focused ultrasound transducers offer a high degree of spatial control, but require long treatment times due to the necessity of dwell times between sonications to protect the rectal wall from thermal damage [28]. Transurethral devices deliver ultrasound from within the prostate gland, enabling continuous sonication and the use of unfocused ultrasound beams for reduced treatment times. The use of broader ultrasound beams, however, requires image guidance to control the extent of treatment accurately. Both ultrasound and MR imaging have been investigated for monitoring heating from interstitial or transurethral ultrasound applicators [11], [29–31].

Integration of ultrasound therapy with magnetic resonance imaging (MRI) enables quantitative images of the spatial temperature distribution to be obtained during sonication for accurate adjustment of treatment to a target volume of tissue. In addition, MRI informs accurate placement of devices relative to soft tissue targets, and can assess the response of tissue and the vasculature post-treatment. This capability of MRI to guide point-by-point coagulation of clinical targets has been demonstrated for breast tumours [32–34] uterine fibroids [35] and bone metastases [36]. MRI-guided ultrasound therapy is evolving towards incorporating temperature measurements obtained during sonication into the treatment to achieve real-time adaptive control over the spatial pattern of heating in tissue. This form of adaptive, closed-loop therapy has been applied for tissue thermal coagulation using intracavitary [18], [37], [38] and externally focused [39–41] transducers, lasers [42], [43], and percutaneous radiofrequency applicators [44].

This manuscript reviews the work conducted by our group over the past 10 years in applying these concepts to the prostate gland. The development of MRI-controlled transurethral ultrasound therapy is described, including the technology and its implementation in the MR imaging environment. The development of algorithms to use MR thermometry as feedback for closed-loop heating and the use of anatomical databases to study the safety and accuracy of this treatment are reviewed. Experimental results obtained with this technology in canines and humans are also described. The overall conclusion from this research is that a transurethral approach provides a quick and efficient method for prostate thermal therapy which, when integrated with MR temperature control, ensures a high level of treatment accuracy.

Technology development

The prototype system developed within our group for MRI-controlled transurethral ultrasound therapy consists of a multi-element ultrasound heating applicator, operated under rotational control within a standard MRI. The multi-element planar transducer emits a directional ultrasound beam producing localised heating in the prostate gland. An MRI-compatible positioning system rotates the applicator to obtain coverage of the targeted region, which can be as large as the entire prostate gland. MR thermometry is performed continuously during ultrasound delivery to measure the spatial temperature distribution in the prostate gland and surrounding tissues. The temperature maps are analysed by the treatment delivery system, and the rotation rate of the applicator, the frequency, and the power output for each element on the transducer are adjusted such that a desired temperature threshold is achieved along the pre-defined target boundary. The system is designed to operate within a clinical closed-bore 1.5 or 3.0 T MR imager. The therapy system is comprised of three main components, described below. The entire system was developed within our research group at Sunnybrook Health Sciences Centre.

Multi-element transurethral ultrasound applicator

The ultrasound applicator (UA) contains a multi-element planar PZT ultrasound transducer with a width of 3.5 mm, and a length determined by the number of elements and the length of each element. Initial devices developed for animal and human evaluation consisted of four or five elements, each 5 mm in length; more elements would be required to sonicate the entire length of the prostate gland in most patients. The transducer is air-backed and mounted in a rigid brass tube with an outer diameter of 6.4 mm (0.25 inch). A prototype four-element UA that was evaluated recently in humans is shown in Figure 1A. An acoustic window is created at the location of the transducer using a bio-compatible 25-µm polyester membrane (Advanced Polymers, Salem, NH, USA). Continuously degassed water (temperature adjustable between 10° and 40°C) is circulated through the device at approximately 300 mL/min during power delivery to remove thermal losses from the transducer and to couple sound into adjacent tissue. The tip of the UA is flexible in order to aid in the insertion of the rigid device into the prostate gland. Electrical power is transmitted from a multi-pin receptacle (Lemo, Ecubiens, Switzerland) at the back end of the applicator to the transducer via a multi-conductor shielded cable (New England Wire and Cable, Lisbon, NH, USA). The UA is able to withstand a sterilisation cycle using ethylene oxide, with no significant change in the acoustic properties of the device.

Figure 1. (A) Photograph of an MRI-compatible transurethral ultrasound applicator used to generate a targeted region of thermal coagulation in the prostate gland. A close-up view of the acosutic window in the inset shows a 3.5 × 20 mm planar multi-element transducer. (B) Photograph of the MRI-compatible rotational positioning system attached to the patient table of a clinical 1.5T MR imager. (C) The concept of transurethral ultrasound therapy is shown with multiple collimated high-intensity ultrasound beams heating a localized region of the prostate gland. Independent control over power and frequency to each element combined with rotation of the ulltrasound applicator enables generation of an arbitrary pattern of thermal damage in the prostate. Multiple MR images are acquired in the planes of the transducers (dotted lines) continuously during treatment to measure the spatial heating pattern in the prostate and control the delivery of ultrasound to a targeted volume.

In addition to the UA, an endorectal cooling device (ECD) is inserted into the rectum to protect that tissue from thermal damage during treatment. The device enables temperature-controlled water circulation at a constant desired temperature. To eliminate flow artefacts in the prostate gland, the circulating water is doped with 5 mM MnCl₂ to shorten T2 sufficiently to remove ECD signal in the images during MR thermometry. The portion of the ECD adjacent to the prostate is constructed of a thin polyester membrane to avoid absorption of ultrasound energy which could result in local heating at the interface between the inner rectal wall and the ECD. The ECD is necessary in cases where ultrasound heating is required in the posterior portion of the prostate gland because the separation between the prostate and the rectum is only a few mm.

Rotational positioning system

The treatment delivery system has the capability to confine heating to angular sectors of the prostate gland through device rotation during treatment in a clinical MR imager. This feature of simultaneous motion and imaging is a central aspect of the feedback control for this technology. The rotation of the ultrasound applicator treatment is slow and continuous (average ∼15°/min; instantaneous from 0 to 120°/min), necessitating precise rotational control to adjust the depth of heating [45]. To achieve these requirements, an MRI-compatible rotary motor was developed, comprised of a custom-machined bearing constructed primarily of plastic, glass and ceramics, two piezoceramic actuators made of non-ferromagnetic components (HR2, Nanomotion, Yokneam, Israel), and an optical encoder assembly (Model LIA-20, Numerik Jena, Jena, Germany). The motor and encoder cables are passed into the magnet room through a grounded connection panel and filtered connectors. The motor and encoder signals are sinusoidal and below 100 kHz avoiding radiofrequency (RF) interference with the MR imager. In addition to rotational control, an MRI-compatible manual positioning system with four degrees of freedom has been constructed to hold the UA in a position aligned with the urethra, and to enable precise advancement or retraction of the device to position the transducer at the desired location within the prostate gland. The positioning system is attached to the patient table during therapy, and enables the rotary motor to be fixed in an arbitrary position and angle relative to the target. The system is designed to be placed between the legs of a human subject. Figure 1B illustrates the positioning system and rotary motor attached to a clinical 1.5 T MR imager (Signa, GE Healthcare, USA).

RF electronics and computer interface

The research treatment delivery system consists of five independent electronic channels capable of delivering high power (≤50 W, 1–10 MHz) RF signals to the UA. Each channel consists of a signal generator (WE7121, Yokogawa, Tokyo, Japan), RF amplifier (NP2910, NP Technologies, Newbury Park, CA, USA), bi-directional coupler (Pulsar Microwave, Clifton, NJ, USA) and custom-made matching network. Each amplifier also contains a custom-made, high-power, low-pass filter to suppress high-order harmonics which might cause RF interference with the MR imaging system. Electrical signals are transmitted via ∼15 m of coaxial cable through a grounded penetration panel into the magnet room where they are connected to a shielded box containing five separate matching networks approximately 2 m from the heating applicator. The signals are transmitted the remaining distance to the heating applicators via a flexible multi-coaxial cable to facilitate applicator handling and rotation. The flexible cable is kept short to minimise resistive losses along its length, especially at higher frequencies.

Two computers are used to control the treatment. The first computer is the hardware interface (HI) which controls the rotation of the motor, and power delivered to each channel. In addition, this computer is capable of measuring the forward and reflected power on each channel, as well as the pressure and flow of the water in the cooling circuits. The second computer is interfaced to the MR imager through an Ethernet connection and acts as the treatment delivery interface (TDI). K-space data are acquired after every image acquisition using a real-time data server on the MR imager [46]. These raw data are reconstructed into images by the TDI and the spatial temperature distribution is calculated for the required number of slices and coils. Typical operating parameters for our animal and human studies are four surface coils and five imaging slices, resulting in 20 MR images that are processed by the TDI. The image acquisition time is typically 5 s, and the data are acquired and reconstructed in approximately 0.5 s. The HI and TDI communicate with each other through a dedicated serial connection between the two computers. The TDI analyses the temperature along the direction of the ultrasound beam during treatment and determines the appropriate adjustments to the rotation rate, power and frequency of each element based on a predefined algorithm in the software [45].

Review of technology development

Our experience confirms that MRI-compatible treatment delivery systems are feasible and can achieve robust motion (in particular rotation) of interventional devices in a closed-bore MRI system simultaneous with imaging. In addition, transurethral ultrasound applicators can be constructed from materials that produce little distortion of the magnetic field. In contrast to other MRI-guided focused ultrasound systems based on externally located transducers [47–49], this minimally invasive approach does not require a dedicated patient table, involves single-use applicators that are low-cost and simple to fabricate, and produces short treatment times due to the continuous sonication approach [50]. These characteristics address many of the hurdles preventing MRI-guided focused ultrasound systems from achieving wider clinical acceptance. The main focus of our investigations has been to test the ability of less focused ultrasound sources in achieving good treatment accuracy.

Device localisation and MR thermometry

Integral to this technology is the use of MR imaging for accurate positioning of the transducers within the prostate gland, measurement of the spatial temperature distribution, and evaluation of the region of thermal damage in the prostate gland.

Device localisation

Precise localisation of the transducer elements within the prostate gland is essential, otherwise undesired exposure to the base or apex of the prostate could result in damage to the urinary sphincters. Insertion of the UA into the prostatic urethra is performed manually by a urologist, without image guidance. This protocol was established in our initial animal studies where it became evident that linear insertion of the UA into the prostatic urethra under motorised or robotic control was not practical. The skilled manipulations that are easily achieved by a trained urologist are difficult to replicate mechanically, and the UA can be inserted into the prostatic urethra in less than a minute within 20–30 mm of the intended location without any form of image guidance.

Once inserted into the prostatic urethra, the UA is attached to the rotational positioning system. Since the UA protrudes out of the penis at an arbitrary angle (approximately 15–20 degrees from horizontal), the positioning system is designed to be articulated to the same angle in order to attach to the UA without applying any torque to it. Two sets of high-resolution, T2-weighted images are acquired to determine the orientation of the UA and location of the transducers in the prostate gland. Both images are acquired with no water flow such that a strong MR signal is produced from the device's acoustic window. The first image is acquired along the UA, transverse to the face of the transducer and enables visualisation of the acoustic window and the transducer. Based on this image, the location of the transducer relative to the prostate gland can be determined. Manual advancement or retraction of the UA to the desired position is performed using a leadscrew stage integrated into the positioning system that is co-linear with the UA. The second image is acquired transverse to the device at the location of the ultrasound transducer in order to determine the angular orientation of the device. Examples are shown in Figures 2A and 2B. Accurate positioning of the UA can be achieved in only a few manipulations and once the 3D position of the ultrasound elements and the direction of the ultrasound beam are known, treatment planning can begin. The uncertainty in angular and axial alignment is typically ±3° and ±1 mm respectively. Any remaining angular misalignment can be corrected based on the spatial heating pattern measured with MR thermometry at the start of treatment, but errors in axial alignment continue throughout the treatment.

Figure 2. (A) T2-weighted images acquired along the length of the ultrasound applicator show the location of the transducers relative to the prostate gland. Based on these measurements, the device can be advanced or retracted to the desired position within the prostate gland. (B) The angular orientation of the ultrasound applicator is determined by acquiring a T2-weighted image transverse to the device to visualize the flat face of the transducer.

Prostate MR thermometry

MRI-controlled transurethral ultrasound therapy involves continuous delivery of high-intensity ultrasound energy to the prostate gland as the UA is rotated about its axis. The temperature distribution in the prostate gland is measured by subtracting MR phase images acquired during treatment from a reference phase image acquired prior to heating. These temperature measurements control the delivery system; thus, it is imperative that they be accurate in order to achieve accurate spatial control over heating. Many of the challenges that exist with MR thermometry using the PRF subtraction method [51] can be avoided in the prostate gland. The presence of the transurethral UA keeps motion of the prostate gland relatively small during a typical treatment of up to 30 min. The prostate does not have a significant fat component around it. Respiratory motion can result in some periodic temperature artefacts in the abdomen but, in our experience, these are typically under 1°C in the prostate.

There are still a number of factors that must be considered in order to perform accurate MR thermometry in the prostate gland during thermal therapy. The water circulating through the UA and the ECD will produce temperature gradients in the prostate gland if the water is not at body temperature. These temperature gradients invalidate the assumption that the prostate is at core body temperature in the reference image, resulting in errors particularly close to the devices. An illustration of the error that can occur in the prostate gland with a room temperature ECD is shown in Figure 3A. Our approach is to avoid flow in devices for some time prior to the start of treatment, enabling their temperatures to equilibrate with core body temperature. Flow is initiated only after the reference images have been obtained. This method enables visualisation of the cooling of tissues around the rectum during treatment, also shown in Figure 3B. Other treatments that involve a sequential approach to heating, such as transrectal ultrasound therapy, require alternative approaches in order to achieve accurate MR thermometry in the prostate gland.

Figure 3. (A) Heat transfer calculations of the temperature distribution along a line extending from the prostate (P) to the rectum (R) in the presence of rectal cooling (20°C). The location of the temperature profile is the dotted line superimposed on the inset schematic of the prostate gland and rectum. Within minutes, a non-linear temperature gradient develops between the urethra and rectum of approximately 10°C. These tissues would be incorrectly assumed to be at core body temperature (37°C) if the conventional PRF subtraction technique was used for MR thermometry. (B) By initiating cooling of the rectum after the acquisition of the reference image, accurate MR temperature measurements can be obtained depicting the extent of cooling in the rectum and prostate gland, as shown in the axial MR temperature maps obtained during MRI-controlled transurethral ultrasound therapy in a canine prostate.

Temperature error can also arise from bulk motion of the prostate caused by rectal peristalsis, faeces, or gas. These events typically cause irreversible shifts in the measured temperature that result in a stop-treatment condition. The presence of a cooling device in the rectum helps to reduce the possibility of movement and pharmacological agents, such as Buscopan, can be administered to reduce rectal peristalsis. The possibility of temperature error due to motion must be monitored during treatment.

Because MRI-controlled transurethral ultrasound therapy can take up to 30 min, the inherent stability of MR phase measurements must be considered. Drift in the background phase is common on MR imaging systems and can be as large as 1°C/min [52]. Compensation for this background drift in temperature is typically achieved by monitoring the temperature in unheated regions of tissue (muscle, fat) or in reference tubes in the imaging field of view.

Taking all of these factors into account enables accurate measurements of the spatial temperature distribution in the prostate gland during MRI-controlled transurethral ultrasound therapy. Practical attention to detail and monitoring for erroneous temperature measurements must be in place before and during treatment. Using these methods, quantitative prostate MR thermometry has been implemented successfully in preclinical animal models and in human subjects.

Active temperature feedback

Treatment delivery strategies incorporating active temperature feedback control have many advantages over open-loop approaches (treatment plans) because they can compensate for unpredictable spatial and temporal variations in tissue parameters, such as dynamic blood perfusion [53]. Various approaches have been used to develop temperature feedback algorithms for MRI-controlled ultrasound therapies; however, they all rely on adjusting the location, power and/or frequency of the ultrasound field. The most basic approach involves the manual adjustment of these parameters through direct observation of temperature images [10], [54]. Sophisticated feedback controllers have been described for creating areas of uniform temperature or thermal dose, but most applications have been limited to controlling temperatures at a single point, at points distributed along a line (1D), or in a plane (2D) [45], [55–58]. Mougenot et al. [52] were the first to publish in vitro and in vivo heating results using 3D spatial and temporal temperature control with MRI-guided HIFU. Their target volumes, however, were still limited to simple geometries (cubes and spheres) with volumes ≤ 2 cm³. While the observed undertreated volumes were typically small (< 1% of the target volume, on average), over-treated volumes were large (>60% of the target volume, on average).

Burtnyk et al. [50] were the first to describe a temperature feedback control algorithm which automatically adjusted the rate of rotation and the ultrasound power and frequency of a moving energy deposition field from multiple planar transducer elements. This algorithm was tested in a theoretical simulation study of MRI-controlled transurethral prostate ultrasound therapy, and the ability to shape volumetric regions of thermal coagulation to large predefined and realistic 3D target prostate geometries (14 to 60 cm³) was demonstrated. The under- and over-treated volumes reported in this simulation study each remained below 4% of the target prostate volume. The 3D algorithm, described below, was based on prior work by Chopra et al. [45] who described a method to control the heating pattern in a 2D plane transverse to a single ultrasound transducer element.

This 3D feedback control algorithm was developed to produce volumes of thermal coagulation that conform to predefined 3D prostate geometries by using temperature and spatial anatomical measurements to modulate the device rotation rate, ω (°min⁻¹), and the ultrasound power, p_i (W), and frequency, f_i (MHz), of each transducer element, i = 1, … , n. For whole-gland therapy, the target boundary would be defined as the outer prostate boundary; however, sub-volumes within the gland can also be targeted with this algorithm. The goal of the control algorithm is to raise the temperature of the outer surface of the target boundary to a critical temperature that represents the onset of thermal coagulation, T_c (°C), in one complete rotation of the applicator. A secondary objective of the control algorithm is to prevent tissue temperatures within the coagulated volume from exceeding an upper limit, T_h = 90°C, to avoid undesirable effects such as boiling and tissue carbonisation.

3D control algorithm

Figure 4A shows a multi-element transducer positioned inside a continuous 3D target boundary, the prostate. An MR thermometry image centred on each transducer element measures temperature changes created within the target boundary in the plane of rotation of each element. Each element emits a directional ultrasound beam normal to the transducer face producing a collimated heating pattern as illustrated in Figure 4B for a stationary device (ω = 0). Along the direction of heating (the white-dashed arrow in Figure 4B), the temperature rises quickly from T_u (the temperature of the water flowing through the transurethral device), reaching a maximum value, T_max, at about 5–10 mm from the transducer surface, then decreasing and reaching ambient body temperature within a few mm (Figure 4C). The distance from each transducer element, i, to the 3D target boundary along the direction of heating is defined as the target radius, r_i, with temperature Tr_i. The control algorithm modulates ω, p_i and f_i based on the temperature difference between T_c and Tr_i (ΔTr_i = T_c − Tr_i), and the value of r_i. The control parameters are updated after every MR image acquisition (typically 5 s) and determined according to the following equations:where k_ω(r_i) is the rotational gain factor, k_p(r_i) is the power gain factor, and r_f is the target radius dual-frequency threshold (mm).

Figure 4. Typical temperature profile along the direction of heating of an ultrasound transducer element. The target radius (ri) sampled from the 3-D target boundary is used to determine the corresponding element's ultrasound frequency (fi). The difference (ΔTri) between the critical temperature (Tc) and the target radius temperature (Tri) is used to control the element's power (pi) and the applicator rate of rotation (ω). The maximum temperature (Tmaxi) was monitored to ensure that it did not exceed an upper limit (Th) thereby avoiding undesirable effects such as boiling or tissue carbonization.

The rate of rotation is continuous to ensure a continuous pattern of thermal damage and to reduce treatment time. The rate of rotation is inversely proportional to ΔTr_i, such that the rotation speeds up when target radius temperatures increase, reducing local energy deposition. The rotational gain constant, k_ω(r_i), decreases with r_i and is determined through a separate series of simple tuning simulations. The maximum device rotation rate, ω_max, ranges from 60° to 120°/min as a function of r_i, to maintain a continuous pattern of coagulation as ΔTr_i approaches zero and to avoid large angular increments during the finite image acquisition time of the MRI. The minimum device rotation rate, ω_min, is set to 8°/min to limit the maximum treatment time in cases when T_max > T_h occurs frequently.

The ultrasound power of each transducer element is proportional to ΔTr_i such that the rate of heating decreases as Tr_i approaches Tc. The element requiring the minimum ω_i was usually operating at p_max, while the power of the other elements was set to a value between 0 and p_max. The power gain constant, k_p(r_i), is an increasing function of r_i, determined through a separate series of simple tuning simulations. The maximum acoustic power, p_max, is typically set from 10 to 20 W/cm² (time-averaged continuous wave, multiplied by the transducer surface area in cm²) in order to coagulate tissues at large distances from the device (greater than 30 mm using 4 MHz) without having T_max exceed T_h in the intervening tissue. The power of an element is turned off when either the target boundary or maximum temperature exceeds T_c or T_h, respectively, allowing tissues to cool.

One exception to the relationships described above occurs at the beginning of a treatment when the applicator remains stationary (ω = 0) and the ultrasound power of all elements is set to p_max. This enables the generation of a certain amount of heat before initiating device rotation.

The dual-frequency radius threshold, r_f, depends upon the resonant frequencies, f_high and f_low, of the transducer. For example, for devices that resonate at 4.7 and 9.7 MHz, r_f is set to 13 mm, a value determined to provide increased treatment accuracy and to decrease treatment times. In cases where the transducer elements operate at a single frequency, f_low, the value of r_f is set to zero. While it is not described in the control equations, the ultrasound frequency can also be forced to switch from f_low to f_high in predefined sectors to reduce the thermal impact on surrounding anatomy near the prostate [59].

MRI thermometry: Considerations for feedback control

The temperature measurements for feedback control are acquired using MR thermometry which has limited spatial and temporal resolution, and noise-limited temperature accuracy, affecting treatment control and the heating pattern produced in tissue [60], [61]. The uncertainty in temperature, δT (°C), is related to the uncertainty in phase in the MR signal when using the PRF shift method. Ignoring the effects of relaxation and assuming a constant bandwidth and field-of-view and that the pulse sequence timing parameters provide acceptable image acquisition times, the following simplified relationship can be obtained[62]:where (ΔxΔy) represents the in-plane pixel area, Δz the slice thickness and t the image acquisition or controller update time. The relationship is oversimplified in two ways. Large in-plane pixel areas eventually compromise temperature uncertainty because the measured temperature includes areas remote from the intended control point. Long acquisition times reduce δT but eventually introduce sufficient lag to compromise temperature tracking at the target boundary. A study by Chopra et al. [61] determined that the ability to create conformal heating patterns with a rotating transurethral device was compromised mainly by the image acquisition time and temperature uncertainty which should remain less than or equal to 5 s and 2°C (standard deviation), respectively. These parameters are achievable using standard gradient echo sequences on a 1.5 T scanner. Another important consideration in MRI-guided transurethral prostate ultrasound therapy indirectly related to the spatial resolution of the images is potential misalignment of the temperature measurements relative to the true direction of heating. Misalignments where the temperatures are measured ahead of the true ultrasound beam direction are of greatest concern as they can lead to large regions of overtreatment.

This control algorithm, based on temperature measurements at the target boundary, performs better than one based on thermal dose in the context of high temperature ultrasound therapy (>50°C), as outlined by Chopra et al. [61]. Although thermal-dose-based controllers have been implemented for other ultrasound devices, the non-linear behaviour of the thermal dose model at high temperatures makes it extremely sensitive to temperature measurement noise and update times, ultimately compromising its utility in this application [63].

Review of experience

To date, our experience in controlling the shape of heating patterns produced by transurethral ultrasound devices has highlighted the importance of using quantitative temperature feedback. Early open-loop experiments in tissue-mimicking gel phantoms using treatment plans obtained from numerical simulations suffered from poor treatment accuracy and repeatability. The same treatment plan delivered multiple times in gel phantoms resulted in heating patterns with high spatial variability due to slight differences in experimental set-up and device performance. In vivo, the challenges are more significant due to unpredictable spatial and dynamic variation in tissue ultrasound and thermal properties. The use of temperature feedback enables the treatment delivery to be modified according to the subject being heated and the experimental conditions, thus enabling a high level of spatial treatment accuracy to be achieved. It is also important to note that the control algorithm that we have implemented is very simple in nature (proportional or inversely-proportional to a temperature difference) and only considers the temperature along the ultrasound beam of each transducer. This simple approach provides good results because it is stable in the context of the relatively slow heating generated by un-focused and directional ultrasound sources. In the case of externally focused ultrasound, however, heating at the focus is extremely rapid compared to the feedback temperature update rate, necessitating more sophisticated model-based approaches to guarantee stability [41], [55], [57]. Nonetheless, MR thermometry provides 4D temperature data (multiple 2D planes in time) which could be incorporated into our MRI-controlled transurethral prostate therapy system to improve the robustness and accuracy of these treatments.

Treatment planning/strategy

Strategies for treatment delivery were first investigated through numerical simulations which enabled the effect of numerous parameters to be determined in a cost- and time-effective manner. Promising approaches were evaluated experimentally in tissue-mimicking gel phantoms which incorporated the effects of MR imaging and other experimental parameters, yet allowed the acquisition of accurate temperature measurements at any location throughout a large homogeneous medium. To investigate the suitability of different treatment strategies in realistic target geometries, virtual canine or human pelvic anatomical models were developed and incorporated in the numerical simulations and in vitro experiments.

Treatment planning: Pelvic anatomical models

Patient-specific pelvic anatomical models of human prostate cancer patients were created using images obtained from an ongoing pilot human study of MRI-controlled transurethral ultrasound therapy at Sunnybrook Health Sciences Centre. After the insertion of the transurethral device, the structures composing the human pelvis were segmented manually from high-resolution T2-weighted axial images, including the prostate, urethra, rectum, bones and neurovascular bundles. These segmented data, reviewed and edited by an experienced body MRI radiologist, provided the exact geometry of the prostate gland and important surrounding tissues that risk incurring injury during treatment with this technology. Figure 5A shows a 3D reconstruction of human pelvic anatomy based on segmentation of clinical MR images, illustrating the position of the prostate gland relative to the rectum and pelvic bones. It is important to consider the deformations of the gland caused by the insertion of the rigid transurethral ultrasound applicator and rectal cooling device as they can alter the prostate shape and its location relative to other structures. Because the rectal cooling device is considered in the segmentation data, the investigation of various strategies for rectal wall protection is possible to investigate treatment safety in this critical region.

Figure 5. Simulation of a full prostate ultrasound treatment. Pixel size of the temperature map : 2 × 2 mm; Slice thickness: 5 mm; temperature noise: 2°C, Dual frequency treatment: 4.1/13.1 MHz. Acoustic power: 20/10 W. (A) 3D reconstruction of pelvic anatomy from 1 mm-step-segmented data acquired from MR clinical images before treatment. (B), (C) Maximum temperature map after the treatment in prostate and surrounding tissues using 3D reconstructed human pelvic profiles: (B) Side slice located at 10 mm from the center of the prostate, (C) Middle slice which coincides with the center of the gland. (D) Simulation of targeting accuracy in a human prostate after transurethral ultrasound exposures. The targeting accuracy of the simulated system is technically limited by the pixel size of thermal maps, and cannot be more precise than ± 1/2 pixel. In modeling, the calculation and interpolation on prostate and thermal boundaries can lead to sub-resolution targeting errors. When lower than ±1/2 pixel (±1 mm in this case), the interpolated targeting error is considered as negligible and displayed in white. Furthermore, error is considered as an overshoot (OS) when higher than +1/2 pixel (+1 mm) and as an undershoot (US) when lower than −1/2 pixel (−1 mm).

Similarly, a database of segmented canine pelvic anatomy was developed from an ongoing in-vivo study of MRI-controlled transurethral ultrasound therapy, thus enabling the investigation of treatment strategies at the preclinical stage.

Investigating treatment strategies: Numerical simulations and gel phantom experiments

These 3D anatomical databases are useful for investigating different treatment strategies in both numerical simulations and gel phantom experiments. A full description of the numerical simulations is available in previous publications [45], [50]. Briefly, biological tissues (prostate, bone, etc.) are modelled according to the acoustic and thermal properties reported in the literature, with the pelvic models used to delimit spatially the different anatomical structures. The acoustic pressure distribution produced by the multi-element transducer is calculated using an approximation to the Rayleigh-Sommerfeld integral [64], and subsequently used to determine the ultrasound power deposition in tissue [65]. Although organ boundaries are defined, the complex reflections and refraction of acoustic energy at these interfaces are neglected in the calculations. The resulting tissue temperature dynamics are simulated according to Pennes’ bioheat transfer equation [66]. MRI-derived temperature feedback is modelled by spatially averaging the temperature into voxels, temporally averaging the temperature over the image acquisition time and by adding random zero-mean Gaussian noise to the temperature measurement. The control algorithm then uses these temperature data to modulate the device rotation rate and the ultrasound power and frequency of each transducer element. Figures 5B and 5C depict the maximum temperature distribution simulated in prostate tissues and surrounding tissues after a full human prostate ultrasound treatment. Two different temperature maps are displayed: an axial slice located superiorly in the gland (10 mm from the mid-gland) and an axial slice at the middle of the gland. The prostate boundary is represented by a black continuous line which coincides with the target boundary in this full prostate treatment. The boundary of the simulated lesion is displayed with a dashed red line which corresponds to the Tc isotherm.

For the gel phantom experiments, a tissue-mimicking material (Zerdine, CIRS, Clifton, NJ, USA) was used. Prototype devices were inserted at room temperature into a cylindrical container containing the phantom material whose attenuation coefficient at 1 MHz and speed of sound were 0.5 dB/cm and 1540 m/s respectively. Similarly to numerical simulations, pelvic anatomical models were superimposed virtually on the gel medium to delimit individual structures. The experiments were performed using a standard quadrature head coil in a 1.5 T MRI (Signa) and the transurethral device was connected to the MRI-compatible treatment delivery system described earlier. Temperature distributions were calculated using the PRF shift method from MR images acquired approximately every five seconds during treatment delivery. These thermometry images were positioned relative to device features evident on T2-weighted MR images, and were centred on each transducer in multiple planes orthogonal to the applicator. Typical imaging parameters were: FSPGR, 5 slices, TE, 10 ms; TR, 69.2 ms; NEX, 1; FOV, 200 mm; slice thickness, 5 mm; image size, 128 × 64 pixels; bandwidth, 15.63 kHz; flip angle, 30°. The K-space data were zero-padded to 128 × 128 before calculating temperature. These parameters resulted in an image acquisition time of approximately 5 s.

The use of numerical simulations and/or gel phantom experiments have proven to be an effective approach and indispensible tool to determine robust feedback control algorithms, optimal transducer designs, effects of various tissue and imaging parameters, as well as to evaluate potential treatment accuracy and safety in patient-specific anatomical models [9], [15], [27], [45], [50], [61], [67], [68]. Numerical simulations constitute an effective research platform to identify the best strategies for treatment planning and control. In particular, different ultrasound exposure strategies were evaluated by observing the effect of acoustic parameters on treatment time, accuracy and safety [69]. Some simulations studies were also performed to observe the effect of transducer sizes on treatment outcomes [70]. Then, an optimised selection of the best treatment planning can be used during gel phantom experiments for a validation of the concept in vitro.

Treatment evaluation: Targeting accuracy, treatment efficacy and safety

As described previously, the goal of the existing control algorithm is to raise the outer target boundary to a target temperature, T_c. After the treatment, the targeting accuracy is determined by comparing spatially the target boundary to the T_c isotherm as calculated from the cumulative maximum temperature images (records the maximum temperature reached in each voxel throughout the treatment). The targeting accuracy is then quantified by calculating the radial difference (spatial distance) between the T_c isotherm and the target boundary, which is the difference between the black and red lines in Figures 5B and 5C. Figure 5D is a 3D display of the prostate gland. The view includes quantitative data of targeting accuracy displayed on the 3D prostate surface, making it possible to visualise the precision of the temperature feedback controller in different regions of the prostate gland. Our databases of canine and human pelvic models provide realistic anatomical diversity in order to evaluate treatment repeatability and robustness across multiple subjects.

Because prostate and lesion boundaries are defined in terms of the pixels composing thermal maps, the targeting accuracy is limited by the pixel size. Differences between boundaries of less than ±1/2 pixel are not meaningful. In realistic clinical conditions, 2D thermal maps have pixels of 2 mm × 2 mm. Thus a targeting error is considered as an overshoot (OS) only when greater than +1/2 pixel (+1 mm) and as an undershoot (US) only when lower than −1/2pixel (−1 mm). While targeting accuracy provides a good measure of how well the control algorithm shaped the heating pattern to the desired target boundary, it is the volume of this spatial difference which has clinical significance – i.e. what volume of prostate tissue did not reach T_c (undertreatment) and what volume of tissue outside of the prostate exceeded T_c (overtreatment).

As tissue thermal effects are a function of temperature and exposure time, some thermal damage may occur at temperatures below T_c. Treatment safety is thus evaluated using the thermal dose model proposed by Sapareto and Dewey [71]. In this context, treatment safety refers to the thermal impact on the important anatomical structures surrounding the prostate (rectum, urinary sphincters, bones and neurovascular bundles) and is expressed in cumulative equivalent minutes at 43°C (CEM43). Since different tissues exhibit different thermal tolerances, the databases of pelvic anatomical models not only provide actual information on the relative conformation of organs in the pelvis, but inform how the time–temperature data should be interpreted in various spatial locations in and around the virtual prostate boundaries. A critical summary of the thermal tolerances of the important pelvic tissues as well as a complete theoretical analysis on patient safety using numerical simulations is described in a separate publication [59].

In-vivo experience

Animal studies

The canine is a commonly used animal model for prostate therapy studies because the urogenital anatomy is similar to humans, and prostate volume is close to that of humans. We have used this model during the development of MRI-controlled transurethral ultrasound therapy to establish the accuracy and safety of the treatment. All of the animal experiments performed by our group have been approved by the animal care committee at Sunnybrook Health Sciences Centre. For all experiments, the animals are fasted and an enema is performed prior to the experiment to clear the lower GI tract. The animals are intubated and maintained under general anaesthesia for the duration of the experiment through inhalation of 5% isoflurane. The urogenital anatomy of the male dog does not permit insertion of a rigid ultrasound applicator into the prostate gland through the penis, so a perineal urethrostomy is performed to expose the urethra for direct manual insertion of the applicator. The animals are placed supine, head-first in a custom-made cradle placed on the table of the MR system. A custom-built, four-channel phased array pelvic coil has been used to acquire images for these experiments. Two of the coils are placed under the dog on the dorsal surface, and the remaining two coils are placed on the ventral pelvic surface. A fibre-optic temperature sensor (Opsens, Quebec, Canada) is inserted into the rectum to determine the core temperature of the animal.

Three groups of studies have been performed in this model using closed-loop MR temperature control. The first series involved treatment of the prostate gland using a single ultrasound transducer and single imaging plane, referred to as a 2D treatment. The second series involved treatment of the prostate gland using a five element ultrasound applicator, enabling axial control along the device, referred to as a 3D treatment. Both of these studies were performed in an acute animal model. In a third study, a 2D treatment was performed in the prostate gland, and animals were maintained for 48 h to evaluate tissue response after treatment. In all of these studies a target region was chosen within the prostate boundary in order to study the histological spatial transition between coagulated and normal prostate tissue. Different transurethral ultrasound applicators were utilised in these studies, but their designs were similar to that described earlier. The ability to obtain stable MR temperature measurements in the prostate gland was observed consistently in all canine studies, with insignificant artefact from respiration or rectal peristalsis. The canine prostate is surrounded by a large amount of fat, making accurate temperature measurements at the boundary of the gland difficult. In addition, the canine pelvis is smaller than a human's resulting in a shorter distance between the applicator and bone than would be experienced in humans. Nonetheless, all the studies to date (n ≈ 25) have confirmed that accurate regions of prostate tissue can be coagulated using a rotating transurethral ultrasound applicator operated under MR temperature control. Figure 6 shows a sagittal view of the canine pelvis, with the locations of four elements used to generate a pattern of thermal damage within the prostate gland shown. The distribution of maximum temperatures measured in each of the four planes at the completion of treatment shows the ability to elevate the target boundary to a threshold temperature of 55°C under temperature feedback control. This result was obtained using an ultrasound applicator with resonant frequency of 8 MHz, delivering 1.75 W of acoustic power/element.

Figure 6. (A) T2-weighted sagital image through the canine prostate depicting the transurethral UA and the locations of elements selected to deliver high intensity ultrasound to the gland. The distance between the dotted lines represents a 5 mm thick imaging slice placed over each element to capture the spatial heating pattern around the device in the plane of rotation during treatment. (B) Maximum temperature maps acquired in the four slices at the completion of a treatment which consisted of rotating the UA clock-wise, from a starting angle represented by the white line to the red line. Approximately 20 degrees of overlap was implemented in this treatment resulting in a total rotation of 380 degrees. The black line represents the target boundary chosen for each slice, the white line the prostate boundary, and the red line the 55°C isotherm measured in the prostate gland.

The performance of this technology in vivo has been evaluated using two metrics in the animal studies. Targeting accuracy is defined as the difference in the radius of the target boundary and the radius of the isotherm representing the target temperature, T_c, typically 55°C in our studies. The targeting accuracy is determined as the average of the radial difference measured at multiple points along the target boundary with an angular resolution of 1–5°. This metric characterises the performance of the control algorithm for a particular treatment but does not consider the accuracy of the temperature measurements themselves. It is very useful for comparison with numerical simulations, and has enabled us to make improvements to the temperature feedback algorithm over time. Treatment accuracy is defined as the difference in the radius of the target boundary and the boundary of thermal coagulation measured on H&E stained histological sections obtained after sacrifice. This metric evaluates the ability of this technology to coagulate a prescribed target volume, which is perhaps a more clinically relevant metric. Errors in MR thermometry, as well as the performance of the control algorithm contribute to this metric. In order to measure the treatment error, a number of steps are required, each with a contribution to the uncertainty in the measurement. First, histological sections must be obtained in the same location and orientation as the imaging plane of the original target boundary. We have developed methods to perform this tissue sampling [72], but errors in the order of ±2 mm can remain after processing. After the tissue section is obtained, distortions in the prostate gland shape exist between the images and the tissue section and must be corrected. A thin-plate-spine co-registration technique [73] has been implemented to achieve this, using the prostate boundary and location of the ultrasound applicator as control points. This step can contribute to the uncertainty in the measurement of treatment accuracy because the deformation is not based on any tissue biomechanical aspects. In addition, the presence of an ECD in the canine often causes significant distortion of the prostate which can result in registration errors, particularly in the posterior portion. Figure 7 summarises the radius of the 55°C isotherm versus the target boundary radius across multiple canine experiments, as well as the boundary of thermal coagulation versus the target boundary radius. These graphs represent the targeting and treatment accuracies, respectively, where an ideal result would be represented by a line with unity slope. The spread in the points from the line represents the spatial precision of the treatment, approximately ±2 mm in these early studies. These results are slightly worse than the targeting accuracy observed in tissue-mimicking phantom experiments, which is usually ±1 mm, partly due to the fact that our feedback control algorithms have evolved throughout the course of the canine studies completed within our group. Nonetheless, these targeting accuracies are still good in comparison to other prostate cancer therapies.

Figure 7. Top: The targeting accuracy of the temperature feedback controller across multiple canine experiments using either multiple element (left) or single element (right) transducers. The graphs show the relationship between the radius of the 55°C isotherm and the target boundary. Bottom: The treatment accuracy from the same canine experiments shows the relationship between the radius of thermal coagulation and the target boundary. The points on the graph were obtained by analyzing the radii along the target boundary with an angular resolution of 1°. The number of imaging slices and total number of points (N) are indicated for each graph. The dashed lines indicate an error margin of ±2 mm. The graphs show that the temperature controller was able to achieve heating to a desired radius, and that this also resulted in good correlation with the pattern of thermal coagulation in the prostate gland. In addition, the 2- and 3-D temperature control algorithms both performed in a similar fashion.

Human feasibility

Based on the positive results of animal studies, we have recently translated the research system developed for MRI-controlled transurethral ultrasound therapy into human studies. Approval to evaluate the feasibility of MRI-controlled transurethral ultrasound therapy was granted by the Research Ethics Board at Sunnybrook Health Sciences Centre to evaluate the feasibility of performing this treatment in men diagnosed with prostate cancer. The study was conducted in accordance with the Canadian requirements for investigational testing of medical devices developed within health care institutions conducting clinical studies. Eight men with localised, low- to intermediate-risk prostate cancer (stage T1/T2a, Gleason score ≤7 (3 + 4), PSA < 15 ng/mL), who enrolled in the study had a sub-volume of their prostate gland treated with MRI-controlled transurethral ultrasound therapy on the morning of their scheduled radical prostatectomy. Subjects were administered a spinal anaesthetic to remove sensation and motor function from the pelvis and legs, and were placed supine on the MR patient table. A sterilised, single-use transurethral ultrasound applicator was manually inserted into the prostate gland and attached to the rotational positioning system described earlier, located between the subject's legs. A plastic enema tip (Flexi-cuff, EZ-EM, Princeton, NJ) was inserted in the rectum along with a fibre-optic temperature sensor (Opsens, Canada) to remove any gas from the rectum and measure core body temperature, respectively. No endorectal cooling was implemented in this study. MR images were acquired using a clinical external phased-array receive coil. Once the ultrasound applicator and coil elements were in place, the subject was positioned in a closed bore 1.5 T MRI (Signa) for imaging and treatment. The subject's vital signs were monitored throughout the treatment using a MR-compatible monitoring system (Precess, InVivo, Orlando, FL, USA). Device localisation and selection of a target boundary was performed using the methods described earlier, and a 2D treatment was performed using two 5-mm elements operating together to create a 10-mm long ultrasound beam in the prostate. A 10-mm thick imaging slice was prescribed transverse to the device, centred on the two active elements, to measure the spatial heating pattern during treatment using a gradient echo sequence ([TE/TR = 9/40 ms], Field of view [FOV] = 260 mm, 128 × 128 matrix size, 5.2 s acquisition time). The target treatment boundary chosen in this preliminary human study was conservative, maintaining a minimum distance of 6 mm from the prostate boundary to preserve the ability to assess the prostatic capsule histologically for positive margins. A 180° sector aimed at the posterior portion of the prostate was chosen for treatment based on the geometry of the gland once the device was inserted into the body.

The initial results of this study have been very encouraging and confirm that MRI-controlled transurethral ultrasound therapy is feasible in humans. The spinal anaesthetic and transurethral device insertion is well-tolerated, and respiratory motion or motion of the prostate gland, has not had a significant impact on the ability to perform quantitative MR thermometry in the prostate gland during the treatment. The duration of the procedure in the MR imager for the initial cases has been approximately 2 h using our research system, and the ultrasound delivery has lasted for approximately 10 min. Improvements to workflow and the software interface and could reduce the 2 h of magnet time required for treatment by up to half, making implementation of this treatment on clinical scanners very practical. The prostate temperature maps using MR thermometry have been very stable with a temperature uncertainty of approximately ±1°C using the imaging parameters listed above. Figure 8 shows the target boundary chosen for treatment in a single subject, the measured maximum temperature distribution at the end of treatment, and the pattern of thermal damage observed in the prostate gland on whole mount histological sections. As in the previous canine studies, the pattern of thermal damage is continuous and well demarcated from surrounding normal prostate tissue. The transition between the boundary of thermal coagulation (light blue line) and outer boundary of thermal injury (black line) corresponds to the treatment margin with this technology, and can be seen to extend approximately 2 mm in Figure 8C. The shape and extent of the boundary of thermal coagulation agrees very well with the shape of the spatial heating pattern measured with MRI and the target boundary selected for treatment. The study is ongoing and expected to complete recruitment for the first phase at 1.5 T by mid-year.

Figure 8. Results of preliminary human evaluation of MRI-controlled transurethral ultraosund therapy. (A) T2-weighted axial image shows the prostate boundary, and boundary of the target region selected for treatment. (B) The maximum temperature map acquired after treatment depicts successful heating of the target boundary to the desired critical temperature of 55°C. (C) An H&E stained whole-mount section in the prostate gland obtained in the plane of treatment depicts a continuous pattern of thermal damage, with a well-demarcated boundary of thermal coagulation. The transition to the outer boundary of irreversible thermal injury is ≤2 mm, consistent with the findings in canine prostate.

Conclusions

MRI-controlled transurethral ultrasound therapy integrates real-time quantitative temperature measurements in the prostate gland with high-intensity ultrasound delivery to achieve adaptive, closed-loop thermal therapy of the prostate gland. Over the past decade, we have developed the MR-compatible hardware and software interfaces to perform these treatments in a closed-bore MR imager. Our experience suggests that this form of treatment is technically feasible, and compatible with existing MR imaging systems. Temperature-feedback algorithms have been developed to control single and multiple element transducers, and the results in gel phantoms and canines have demonstrated the feasibility of producing accurately targeted volumes of thermal coagulation in the prostate gland with this technology. Patient-specific simulations have been used to study the accuracy of the treatment under realistic prostate geometry and imaging conditions, and to predict the impact of this treatment on surrounding tissues. Initial results of those investigations suggest that important tissues such as the rectum, bladder, bone, and nerves can be spared from thermal damage if appropriate safety measures are taken into account during treatment planning. Finally, recent experience in the translation of this technology into a human study has been encouraging and motivates continued study of this technology as a potential treatment for localised prostate cancer.

Acknowledgements

The authors would like to thank Ilya Kobelevskiy for his assistance in the preparation of figures.

Declaration of interest: The funding for this work has come from a number of sources including the Canadian Institutes for Health Research (CIHR), National Cancer Institute of Canada (NCIC), the Ontario Institute for Cancer Research (OICR), the Terry Fox Foundation, and the Ontario Research Fund (ORF). The technology described in this manuscript has been licensed to Profound Medical Inc. by Sunnybrook Health Sciences Centre. Rajiv Chopra and Michael Bronskill are founders and hold shares in the company. The clinical results presented in this manuscript were obtained using the research system developed at SHSC. Profound Medical did not have any role in the financing, planning, recruitment or analysis of that study. The authors alone are responsible for the content and writing of the paper.