The Bayesian approach was also evaluated with experimental data from two differ-ent measuremdiffer-ent setups. In publicationII, measurements were carried out using a Fabry-Pérot sensor based photoacoustic measurement system developed in the Pho-toacoustic Imaging Group of the University College London [75,199]. In publication III, PAT measurements were done using an LED-based imaging setup developed at the Department of Applied Physics in the University of Eastern Finland.
4.2.1 Fabry-P´erot sensor based PAT setup
In the case of the Fabry-Pérot sensor based setup, the imaged objects were a skeletal leaf phantom and a mouse head. In the phantom measurement, an illumination of an imaged object was done using a laser operating at the wavelength of 1064 nm, whereas for the mouse measurements two optical parametric oscillators were tuned
Figure 4.10: A photograph of the leaf phantom (left image), the contour surface of the reconstructed photoacoustic image of the leaf phantom obtained using the orthogonal sensor (middle image) and the planar sensor (right image). The rows from top to bottom represent the reconstructed image obtained using the Bayesian approach (first row) and time reversal (second row).
to the wavelength of 755 nm. PAT signals were detected with a planar or orthogonal Fabry-Pérot sensor by scanning an area of approximately 10 mm×10 mm on the sensors with a step size of 100µm. In all measurements, the imaged targets were coupled to the sensor using deionized water.
Before reconstructions, nuisance signal components of the measured PA signals were removed by filtering (a bandpass filter with cutoff frequencies between 0.5 and 20 MHz). This filtering was also taken into account in the forward model. The image reconstruction was performed by solving the system of equations (3.13). The compu-tations were done in 3D, and the grid sizes were 274×248×242 and 304×286×240 cubic voxels (voxel side length 50µm) for the leaf and mouse measurements, respec-tively. In the computations, the speed of sound was assumed to be 1488 m/s. The Ornstein-Uhlenbeck prior with the meanηp0 =0, standard deviationσp0 =0.25 and characteristic length scale l = 0.1 mm was used as the prior model. In addition, the noise statistics at each sensor position were determined by calculating the mean and standard deviation from a time frame of the measured PA signals that is sup-posed to contain only noise. Furthermore, a time reversal solution was computed for comparison.
The experiments with the Fabry-Pérot sensor show that reconstructions obtained using the Bayesian approach represents the features of the imaged target (Figures 4.10 and 4.11). That is, the vein-like structure of the leaf and the vasculature of
Figure 4.11: The photoacoustic images of the mouse head obtained using the Bayesian approach (first row) and time reversal (second row). The left image shows the contour surface of the reconstructed image. The three images on the right rep-resent maximum intensity projections along axis directionsx,y, andz.
the mouse head are visible in the reconstructions. In addition, the Bayesian recon-structions are similar with the time reversal reconrecon-structions (Figures 4.10 and 4.11).
However, some differences between the reconstructions obtained with the Bayesian approach and time reversal can be seen, especially in the case of the mouse head. In addition, the Bayesian approach seems to be able to detect structures deeper than the time reversal which is more evident in the case of the leaf phantom. Further-more, the quality of the image reconstructed from the orthogonal sensor appears to be superior to that of the image reconstructed from the planar sensor since it gives a more complete reconstruction of the structure of the imaged target (Figure 4.10).
However, the structures that are close to the sensor appear sharp with both sensors.
4.2.2 LED-based PAT setup
In the LED-based PAT measurements, an imaged object was illuminated with an LED operating at the wavelength of 617 nm. PA signals were acquired using a cir-cular piston ultrasound transducer that was rotated around the target. Different limited view and sparse angle measurement situations were considered. The imag-ing targets were made of plastic microcapillary tubes that were filled with an Indian ink solution, and they were coupled to the sensor using degassed deionized water.
The sensors were modeled and the sensor response was included in the forward operator K(c). The image reconstruction was performed by solving the system of equations (3.13). The computations were conducted in a 14.6×8.2 cm domain that was discretised into 730×410 square pixels. The Ornstein-Uhlenbeck prior with the mean ηp0 = 0.015, standard deviation σp0 = 0.005 and characteristic length scale
Figure 4.12: A schematic picture of the measurement setup and the reconstructed photoacoustic images from sparse angle measurements of three ink (0.1%) filled tubes when the direction of the tube pattern was in at approximately 45◦ angle relative to the light source. From left to right and from top to bottom the separation between the measurement angles and the number of the measurement angles in brakects are 1◦(185), 2◦(93), 5◦(37), 10◦(19), 20◦(10), and 30◦(7).
l = 0.85 mm was used as the prior model. In addition, the mean and standard deviation of the measurement noise was calculated from a time frame preceding the measured PA signals at each sensor position. The reconstructions were computed using the speed of sound that corresponded the temperature of water.
The experiments with the LED-based setup show that the LED-based instrumen-tation can also be utilised in limited view and sparse angle measurement setups. In addition, the experiments confirm that the images reconstructed with the Bayesian approach present the features of the imaged target (Figure 4.12). Furthermore, the results show that the Bayesian approach enables a significant reduction of measure-ment angles without compromising the quality of the reconstructions. That is, the number of measurement angles can be reduced from 185 to 37 without losing the quality of the image. In fact, the tubes can be visually distinguished from the arte-facts even when only 19 measurement angles are used.