1.

Introduction

Heat is an effective tool for treating cancer.^1,2 In radiotherapy and chemotherapy, heat synergistically interacts with either the radiation or the cytotoxic drugs. Hyperthermia, which raises the temperature of a cancer region to 40 to 43°C, is usually applied as adjunctive treatment to improve radiosensitivity or drug delivery. Moreover, as an alternative cancer treatment, thermotherapy heats tissues to even higher temperatures (46 to 70°C) and causes tumor damage through thermoablation. In these thermal treatments, it is necessary to control the temperature in order to either maintain the desired temperature level or minimize damage to surrounding healthy tissues. Therefore, monitoring the temperature distribution around the tumor region deep in tissue in real time is critical to ensure effective treatment of the target tumor region and avoid or minimize collateral damage.

Several noninvasive methods have been used to measure temperature. Infrared thermography is a real-time method with good accuracy, but it measures only superficial temperature.³ Ultrasound imaging provides good spatial resolution and penetration depth, but it has low temperature sensitivity.⁴ Magnetic resonance imaging offers high resolution, deep penetration, and high sensitivity, yet it is expensive, bulky, and slow.^5,6

Photoacoustic (PA) tomography is capable of structural and functional imaging.^7–9 Several reports have detailed the capability of PA techniques for temperature sensing as well.^10–15 The temperature-sensing capability is based on the linear relationship between the amplitude of the PA signal from the medium and the temperature. Combined with the reported results of imaging tumor angiogenesis and monitoring cancer therapy,^16,17 by using blood as a natural contrast, PA tomography can be a promising tool to monitor the temperature within the range for hyperthermia therapy. However, so far these studies have focused on temperature sensing at single positions, i.e., only the temperature change over the time at a given position has been monitored. In other words, measuring spatial temperature distribution over a large region using PA techniques has not been reported.

We present in this paper our experimental results on imaging temperature distributions in tissue-mimicking phantom using PA tomography. We first calibrated our PA imaging system onsite by measuring both the temperature $T$ at a spot in the tissue using a thermocouple and the PA signal from that spot. The resultant linear PA–$T$ relationship was used to translate the PA image of the tissue into a temperature distribution. Our results demonstrate that it is possible to calibrate the system for a greater temperature range in amplitude ($>10°\mathrm{C}$) by single-point measurements over a small temperature range. Since natural blood provides the PA contrast, the calibration can be accomplished on or near the surface of the body without inserting a thermocouple deep into the body, an advantage in clinical application. In fact, with some potential loss of accuracy, calibration can be done entirely noninvasively, by imaging superficial blood vessels. Moreover, such in vivo calibration can be achieved in situ, which is more accurate than ex vivo calibration.

2.

Methods and Materials

2.3.

Sample Preparation and Imaging

The sample was prepared using fresh chicken breast tissue. The surface of a piece of chicken tissue was stained with black ink to provide optical contrast. Another piece of chicken tissue was then placed atop the stained surface. A heating plate was inserted beneath a part of the bottom chicken tissue. Therefore, only part of the sample was heated directly, creating a temperature gradient across the whole sample. Two thermocouples were inserted between the two pieces of chicken tissue to measure the temperature of the ink layer at two positions during the experiment. Black ink instead of blood was used to provide optical contrast, simply because blood may dry out during the experiment. However, both ink solution and bovine blood were previously studied for PA temperature sensing.^14,18 This substitution should not affect the general conclusions.

The experimental setup is illustrated in Fig. 1(a). The ultrasonic transducer was immersed in water inside the water tank. The tank had an open bottom that was sealed with a thin polyethylene membrane. The sample was placed outside the tank below the membrane, and the ultrasonic coupling was secured by coupling gel between the sample and the membrane. The transducer, as well as the optical assembly, raster scanned in the $xy$ planes parallel to the ink surface. The step sizes were 0.2 mm in the $x$-direction and 0.75 mm in the $y$-direction. (Note that $y$ points out of the paper.) Each complete scan took about 10 min and generated a three-dimensional PA image. A maximum amplitude projection (MAP) image projected along the depth direction is produced.

Fig. 1

(a) Schematic of the experimental setup. (b) Representative B-scan photoacoustic (PA) image of the sample acquired at the baseline (i.e., room) temperature (22°C) before heating the sample. It was used as a reference to normalize all the subsequent PA images. P1 and P2 indicate the positions of the two thermocouples. TH, thermocouple and UT, ultrasonic transducer.

First, a scan was performed when the whole sample was at the baseline temperature, $T_0$, and a two-dimensional (2-D) image was acquired as a reference. Then the heating plate was turned on to increase the temperature. Another scan was performed when the temperature readings from the two thermocouples reached steady state, suggesting the temperature distribution in the sample had become stable. The same procedure was repeated several times with the heating plate set to different temperatures. The readings from one of the thermocouples and the PA signal amplitudes at the corresponding position were used to calculate the PA–$T$ curve. The curve was then used to calculate the temperature distribution of the ink layer from the 2-D PA images.

3.

Results and Discussion

A PA B-scan image [Fig. 1(b)] showing a vertical $x–z$ cross-section of the sample was acquired by scanning the system in the $x$-direction. Although the top surface of the chicken breast tissue was barely visible, the ink layer about 0.5 cm below the tissue surface generated strong PA signals. The PA signal amplitude was not uniform along the ink layer even though the whole sample was at the same baseline temperature. This variation was attributed to multiple factors: the ink stained on the chicken tissue was nonuniform, creating spatial fluctuation in optical absorption coefficient; the top layer of tissue had a varying thickness, causing variation in optical fluence at the ink layer; the chicken tissue was inhomogeneous, again making the fluence position-dependent. Fortunately, normalization by the reference PA amplitude in Eq. (4) removes the effects of these factors.

Multiple B-scan images similar to Fig. 1(b) were acquired as the system raster-scanned the sample. An MAP image projected along the depth direction was then generated to show a 2-D image of the ink layer in the $xy$ planes. In the B-scan image in Fig. 1(b), the positions of the thermocouples are indicated by arrows. Although invisible in this image, the thermocouples did show up in other B-scan images, helping us to locate the exact positions. PA signal amplitude at each of these two positions was averaged over a $2.00\times 2.25\mathrm{mm}$ ($x\times y$) window in the 2-D image.

The normalized PA amplitude at each of the two positions was plotted in Fig. 2 as a function of the temperature measured by the thermocouples. The crosses show the measurements taken at position 1, the hot end. The dashed line is a linear fit of this measurement:

Fig. 2

Normalized PA amplitude versus temperature measured by the thermocouples. Blue crosses and red dots show the measurements at positions 1 and 2, respectively. The blue dashed line is the linear fit of the measurements at position 1 through $\widehat{p}=0.036T+0.21$ and the red dotted line is the fit at position 2 through $\widehat{p}=0.037T+0.21$.

Obviously, measuring over a large range of temperature will yield more accurate fitting. However, as shown in Fig. 2 by the red dots and the dotted line, the measurement over a smaller temperature range at position 2, the cold end, gave

Figure 3 shows maps of temperature acquired sequentially at two of the seven heating plate temperatures. The distributions were calculated from the 2-D PA images using Eq. (4), where $b$ was determined by Eq. (6) from the cold end. The map was averaged over a $1.00\times 2.25\mathrm{mm}$ ($x\times y$) window at the cost of spatial resolution. Since the scanning had different step sizes in the $x$ and $y$-directions, we chose a smaller window size in the $x$-direction to maintain a better resolution. The dotted lines in these figures indicate the edge of the heating plate underneath the sample. The hot side of the sample was on the heating plate and the cold side was not. The heating plate was tilted intentionally to misalign the temperature gradient with both scanning directions. The areas occupied by the thermocouples are blanked because in these areas the sample was slightly deformed and the PA signals were not from the ink layer. The temperature distribution was expected showing a decrease from the heated part to the nonheated part of the chicken tissue. Along the edge of the heating plate, the temperature was higher in the middle, which was possibly due to less heat exchange between the sample and the environment. The thermocouple readings at the selected spots are shown in this figure too.

Fig. 3

Temperature maps acquired by translating PA amplitude into temperature at (a) a representative low heating plate temperature (35°C) and (b) a representative high heating plate temperature (50°C). The white areas in the images indicate the positions of the thermocouples, whose readings are labeled. The dashed lines in the images show the edge of the heating plate underneath the sample. The hot (lower right) side of the sample was on the heating plate while the cold side was not. Each pixel in the images was averaged over a $1.00\times 2.25$ ($x\times y$) mm area, which is considered here as the spatial resolution for temperature mapping.

Although the measurements at position 2 (cold end), as shown in Fig. 2, were used for calibration [Eq. (6)], the seven measurements at position 1 (hot end) were then used to estimate the accuracy found to be $-0.8\pm 0.4°\mathrm{C}$ ($\mathrm{mean}\pm \text{standard error}$). The accuracy can be further improved by signal averaging. Since all the PA signals were normalized by the baseline, simply averaging the baseline signal could greatly improve the accuracy of the calibrated PA–$T$ curve, therefore improve the overall accuracy.

Our imaging system has good spatial resolution.¹⁹ But the spatial resolutions of the temperature map were limited by the size of the window we used for spatial averaging and were as fine as 1.00 mm along the $x$-axis and 2.25 mm along the $y$-axis. Theoretically, the resolutions should be the same in both axes. However, in our experiment, we were limited by the scanning speed of the system. We chose large step sizes, especially in the $y$-direction, to complete the scan before the chicken tissue dehydrated and changed its properties. The spatial resolution can be improved by reducing the step size of the scan with a faster scanning system. With faster scanning, the scan can be repeated at the same temperature distribution and then averaged to improve the signal-to-noise ratio. This not only improves the accuracy of the PA–$T$ curve, but also improves the spatial resolution.

Since our method measures blood in the temperature range of hyperthermia, heterogeneity in the Grueneisen parameter of the surrounding tissue would not significantly change the value of the constant $b$. Further, a study on canine liver showed that the optical properties did not change during moderate heating.¹² Therefore, the optical heterogeneity is taken into account by the normalization performed in our method.

Our method works within a moderate temperature range where the linearity assumption holds true. Therefore, the primary application is in hyperthermia rather than thermal ablation.

4.

Conclusions

In summary, we investigated the possibility of imaging temperature distribution in deep tissue using the PA technique. The linear relationship between the PA signal and the temperature, the PA–$T$ curve, can be determined through measurement of a small region over a small temperature range. The temperature distribution within the temperature range of hyperthermia can be calculated from the PA image using the PA–$T$ curve. We demonstrated that the method provides a good accuracy and spatial resolution. PA temperature mapping has the capability to monitor the temperature, both temporally and spatially, in deep tissue in real time during thermal treatment of cancer such as hyperthermia or thermotherapy.