Introduction

MR-guided LITT is clinically used to treat a variety of brain diseases, such as glioblastoma and epilepsy [1], [2]. Treatment duration can be of several minutes, depending on the targeted ablation size and laser configuration. Excessive temperature increase may result in alteration of healthy structures and/or lead to incomplete treatment (modification of light penetration in tissue due to charring near the laser probe) and produce thermometry artifacts [3]. We propose to combine real-time 3D thermometry with an automatic regulation algorithm to adjust dynamically the delivered laser power to force temperature to follow a predefined profile.

Material & Methods

Real-time MRI thermometry pipeline: Multi-slice single-shot echo planar images [4] were acquired in coronal orientation every second (6 slices/s, 800 repetitions) on a 1.5 T clinical scanner (Avanto, Siemens Healthineers) with the following parameters: TE=18 ms, FOV=180x180 mm², 3 mm thickness (1.4x1.4x3 mm³ voxel), FA = 60°, GRAPPA acceleration=2, bandwidth/pixel=1446 Hz. The flex and two spines coils surrounded the sample, with 12 channels used during real-time image reconstruction. Raw data were streamed to the reconstruction and thermometry pipeline [5] and temperature images were displayed in real-time on our MR-guided interventional application (Certis Solution, Certis Therapeutics, France). A real-time temporal Kalman filter (with parameters σ²_measurement=5 and σ²_process=1 estimated using a combination of temporal evolution of temperature in several pixels of the heated area over several slices and assuming an autoregressive 4th order process) was applied to denoise each temporal temperature pixel profile from all slices.

LITT device: A 400 µm optic fiber ended with a glass diffuser tip of 1 cm was inserted into the test gel (3% gelatin) and connected to a laser diode source (976 nm wavelength, maximum output power of 27 W), located outside the Faraday cage. The laser unit (ALPhANOV, Talence, France) was interfaced with the Gadgetron to dynamically control and update its output power by software means.

Regulation algorithm: A dedicated processing was implemented in the thermometry pipeline to compute and update the laser power on-the-fly. The controller is a proportional-integral-derivative (PID) algorithm incorporating the Bio-Heat Transfer Equation, as previously proposed [6]–[8]. Input parameters are 1- a temperature-time profile, 2- absorption (α) and diffusion (D) coefficients of the targeted medium and 3- a parameter related to the response time of the PID controller. A 3x3x3 rectangular ROI was centered on the hottest pixel defined from the test shot (see below) to control the energy deposition.

Initial low power test shot: To calibrate the thermal load deposition and the threated area, a CW laser emission (3W output power) was applied during 20 seconds to detect the heated region on temperature images by finding the hottest point averaged over the temporal dimension. The temperature evolution at the hottest point was fit with the following equation [5] to estimate α [°C/s/W]:

$$T(t) = \left\{\begin{array}{lll}0&\mbox{if } t ≤ t_0\\
αPτ\ln[\frac{t-t_0+τ}{τ}]&\mbox{if } t_0 < t ≤ t_1&\qquad &\text{(Eq.1)}\\
αPτ\ln[\frac{t-t_0+τ}{t-t_1+τ}&\mbox{if } t < t_1 \end{array} \right. $$

where t₀ and t₁ are the start and end times of laser emission, P [W] the laser power and τ [s] the characteristic diffusion time of the system. At each time point after the end of the laser emission (cooling phase), the temperature maps crossing the previously selected pixel was fitted with a 2D Gaussian function to retrieve the full width at half maximum of the temperature distribution. These values were plotted as a function of time and the resulting curve fitted to a linear function whose slope is equal to 2D [9].

Automatic regulation: The MR-thermometry sequence was run during 800 s and the regulation algorithm automatically adjusted the laser output power for approximately 13 minutes, with online display of temperature images, predefined profile and measured temperature curves at the focal point.

Results

Figure 1 shows the maximal temperature measured during the initial low power test shot, without and with Kalman filtering. Temporal mean ± standard deviation of temperature measured over the 20 first acquisitions in the time series was 0.29±0.39°C, and 0.28±0.17°C without and with Kalman filtering, respectively. Please note that no significant time delay was introduced by this filter during heating. Figure 2a shows the filtered temperature curved of Figure 1 with the resulting fit (solid red lines) to equation 1 to derive α (0.36 °C/s/W). Figure 2b shows the result of the full width at half maximum versus time during the cooling period, together with the linear fit to determine D (0.2 mm²/s). Figure 3 shows an example of the automatic temperature regulation for a predefined heating profile with 3 steps of 5°C, 10°C and 15°C, and its corresponding temperature maps (b, c d, respectively). Figure 4 displays the difference between measured and target temperature curves over the laser emission duration. The mean difference is -0.1°C with confidence interval (RMSE) of 0.47°C, showing the excellent stability of the proposed technique over long duration (>10 minutes) heating.

Conclusion & Discussion

The proposed regulation algorithm provides precise control of tissue temperature with a mean error between target and experimental temperature below 0.5°C. The Kalman filtering allows reducing temperature noise without adding delay that may compromise the stability of the regulation algorithm. This method offers a stable approach, even for ablation time longer than 10 minutes.

Acknowledgements

No acknowledgement found.