2.1.

Optics and Light Delivery

As the irradiation source, the tunable dye laser (Cobra, Sirah Laser-und Plasmatechnik GmbH, Germany) was pumped by the Nd:YLF laser (INNOSLAB, Edgewave GmbH, Germany). The Nd:YLF pump laser had a pulse duration of $7\mathrm{ns}$ and a pulse energy of $12\mathrm{mJ}$ at $523\mathrm{nm}$ . The continuous optical pumping from the diode stacks in this Q-switched Nd:YLF laser provided the flexibility of external triggering on demand at rates up to $1\mathrm{kHz}$ without compromising the pulse energy. This feature offered a significant advantage over flashlamp-pumped Q-switched lasers, which are typically maintained at a fixed low pulse repetition rate (e.g., $10\mathrm{Hz}$ ). Rhodamine 6G laser dye was used to enable a peak output of $2\mathrm{mJ}$ per pulse with a pulse width of $7\mathrm{ns}$ at the $570\text{-}\mathrm{nm}$ wavelength from the dye laser, with a $40\text{-}\mathrm{nm}$ tuning range. For our imaging experiments, we used the peak $570\text{-}\mathrm{nm}$ wavelength. This wavelength also corresponds to an isosbestic point where oxy- and deoxy-hemoglobin have equal molar extinction coefficients.

Proper delivery of laser light into biological tissues for photoacoustic excitation is crucial to achieving a high signal-to-noise ratio (SNR) in photoacoustic imaging. The light delivery system was designed to provide a compact photoacoustic imaging device with sufficient SNR and robust performance. The dye laser output was coupled into a $0.6\text{-}\mathrm{mm}$ -core-diameter multimode optical fiber and collimated by a fiber collimator at the output end of the fiber. A free-space optic setup, integrated with the ultrasound array, was used to guide light further (Fig. 1). The collimated light beam, of $\sim 6\mathrm{mm}$ diam and $\sim 1.2\mathrm{mJ}$ energy, was split into two beams by a 50:50 nonpolarizing beamsplitter. The two beams were reflected by mirrors toward two cylindrical lenses and coupled into a plastic slab. The $6\text{-}\mathrm{mm}$ thick plastic slab, with a hollow cylindrical core for the ultrasound probe, was polished for light transmission. During experiments, the cylindrical space was filled with water and sealed by a piece of thin low-density polyethylene (LDPE) film fixed by an o-ring. In total, $\sim 80\%$ of the light energy reached the surface of the film. Because of finite fiber aperture the final optical illumination patterns on the skin surface were thick-line shaped, as shown in Fig. 1. The length and width of each illumination area were 6 and $3\mathrm{mm}$ , respectively.

Dark-field laser pulse illumination was achieved through fine tuning the mirrors and cylindrical lenses, reducing the photoacoustic signals from the superficial paraxial area. However, a large dark-field area may reduce the optical fluence reaching the targeted area. The optimal illumination radius was estimated to be $7\mathrm{mm}$ using the concept of effective attenuation coefficient,¹³ the exponential decay rate of fluence far from the source, with typical tissue parameters. Consequently, leaving a $\sim 1\text{-}\mathrm{mm}$ width right below the array elements as the dark field gives approximately the best performance. In practice, the mirrors and cylindrical lenses were finely tuned to optimize the SNR.

The optical fluence on the skin surface was estimated to be $\sim 2\mathrm{mJ}∕{\mathrm{cm}}^2$ per pulse, well below the ANSI recommended maximum permissible exposure (MPE) of $20\mathrm{mJ}∕{\mathrm{cm}}^2$ for a single pulse. We acquired data in $1\mathrm{s}$ —50 frames for realtime B-scan imaging or 166 frames for 3-D imaging; the time-averaged light intensity during this $1\mathrm{s}$ was $\sim 600\mathrm{mW}∕{\mathrm{cm}}^2$ (the total illuminated surface area for one 3-D image was $1.2{\mathrm{cm}}^2$ due to the mechanical scanning), also below the ANSI recommended MPE calculated as $1.1t^{0.25}\mathrm{W}∕{\mathrm{cm}}^2$ ( $t$ in seconds).¹⁴ For prolonged illumination, the ANSI-recommended MPE for average light intensity would be lower. However, for prolonged illumination during B-scan imaging, we can either pause a few seconds between acquisitions or slow down the frame rate. We also expect to reduce the delivered energy with improved SNR by system optimization. The ANSI safety limit for this pulse width region is dominantly based on the thermal mechanism; thus, our compliance to the ANSI standards guarantees no thermal damage to the tissue.

2.2.

Ultrasound Array and Beamforming

We used a unique $30\text{-}\mathrm{MHz}$ ultrasound linear array fabricated from a 2-2-piezocomposite by the NIH Resource Center for Medical Ultrasonic Transducer Technology at the University of Southern California.¹⁵ The array had 48 elements (of dimensions $82\mu \mathrm{m}\times 2\mathrm{mm}$ ) with $100\text{-}\mu \mathrm{m}$ spacing. The dimension of the element in the elevation direction was $2\mathrm{mm}$ , and the elements were focused in this direction with a fixed focal length of $8.2\mathrm{mm}$ , which provides a resolution of $200\mu \mathrm{m}$ in the elevation direction within the $\sim 3.5\mathrm{mm}$ focal zone. The pulse-echo insertion loss and element cross-talk were 19.1 and $-25\mathrm{dB}$ , respectively. The mean fractional bandwidth was 50% for pulse-echo operation, which translates to $\sim 70\%$ for receiving-only operation, as used in our present photoacoustic imaging system.

Althrough ultrasound beamforming traditionally has used dedicated hardware,^{16, 17} we instead used multicore processors (Dell Precision 490 with two $2.66\text{-}\mathrm{GHz}$ Quad core Xeon processors), which allows off-the-shelf personal computers to perform the task and offers programming flexibility. Microsoft Visual Studio 2005 Professional Edition and Microsoft XNA Game Studio in the Visual Studio C# 2005 Express Edition environment were used to develop the software for dynamic receive beamforming and display. Details on implementation of multithreaded parallel programming and GPU-based scan conversion and display in this software beamforming can be found elsewhere.¹²

2.3.

Data Acquisition and Volume Imaging

Photoacoustic signals picked up by the ultrasound array were amplified by a custom-built radio-frequency (RF) board with a $33–73\mathrm{dB}$ variable gain and were down-multiplexed to eight channels, which were digitized at 125 megasamples per second using a $14\text{-}\text{bit}$ 8-channel PCI DAQ card (Octopus CompuScope 8389, GaGe Applied Systems, USA). The card was used as the master clock for the entire system and programmed to send trigger signals to the multiplexer control and laser. The repetition rate was set at $1\mathrm{kHz}$ , which was the highest rate that the laser could work without degradation of pulse energy.

We used linear scanning to achieve 3-D imaging. During the scanning, the array translated linearly over the skin surface, so that the B-scan imaging planes were all parallel to each other. This was accomplished by mounting the light delivery system and the ultrasound array in a linear motion actuator (KR20, THK Co. Ltd., Japan). A bipolar stepper motor (4118S, Lin Engineering, USA) controlled by a microstep stepper motor controller (BC2D15, Peter Norberg Consulting, Inc., USA) was used to drive the linear motion actuator. The scanning system provided sufficient precision $\left(20\mu \mathrm{m}\right)$ for our use. Six laser shots were needed to obtain one B-scan image because of the 6:1 down-multiplexing in our data acquisition. To produce one 3-D image, 166 B-scan frames were acquired in $996\mathrm{ms}$ , corresponding to 996 laser shots at a $1\text{-}\mathrm{kHz}$ repetition rate. During the data acquisition, the array scanned continuously at a constant speed $(10\mathrm{mm}∕\mathrm{s})$ . The speed was set so that the distance the array traveled during each B-scan time period was $60\mu \mathrm{m}$ , less than the ultrasonic focus in the elevational direction $\left(200\mu \mathrm{m}\right)$ . Although the linear motion actuator was fixed on an optical table for scanning, handheld operation was also possible. More user-friendly handheld operation for daily clinical use can be achieved by shrinking the size of the light delivery and scanning systems through custom manufacturing.

While real-time B-scan imaging was demonstrated at $50\mathrm{fps}$ ,¹² postbeamforming after data acquisition reached $83\mathrm{fps}$ . This higher speed was due to less hardware communication. In postbeamforming, a B-scan movie was first played, and an MAP of the acquired 3-D image was then displayed for preview immediately. In total, $\sim 3\mathrm{s}$ was needed for a user to view an MAP image, representing the fastest speed among reported photoacoustic imaging systems. The user might choose to either replay the B-scan movie or display a contrast-enhanced MAP image processed by a Dynamic-link Library written in MATLAB (Math Works, Inc., USA). All these operations could be done online by simply clicking corresponding buttons on a graphic user interface generated by the C# program.