Abstract
Freehand 3D photoacoustic tomography (PAT) typically requires expensive external tracking systems to register 2D image slices into a 3D volume. We present "Tattoo Tomography," a sensorless tracking approach using a simple trident-shaped optical pattern applied to the skin. The pattern is imaged natively by the PA probe, allowing for real-time 3D pose estimation through trigonometric analysis. Our method achieves a fiducial registration error of 0.63 mm, surpassing commercial optical trackers and offering a low-cost, robust solution for clinical 3D-PAT.
Introduction
Photoacoustic Tomography (PAT) is a hybrid imaging modality that combines the high contrast of optical absorption with the deep penetration of ultrasound. While 3D-PAT offers significant clinical value for visualizing vascular structures and functional hemodynamics, most clinical systems utilize 2D handheld probes. Reconstructing a 3D volume from these 2D slices requires precise knowledge of the probe's spatial trajectory.
Current tracking solutions, such as optical or electromagnetic trackers, introduce significant overhead: they are expensive, require complex calibration, and often suffer from line-of-sight obstructions. Sensorless methods like speckle decorrelation, while effective in ultrasound, fail in PAT due to the lack of acoustic speckle. To address these challenges, we propose "Tattoo Tomography," which uses the photoacoustic signal of a surface-applied pattern for intrinsic tracking.
Methodology
The core of the proposed system is a trident-shaped optical pattern—an isosceles triangle with a central bisecting line—printed on transparent foil. This "tattoo" is printed with cyan ink, which has high optical absorption at 750 nm, making it clearly visible to the PAT system.
As the 2D probe moves across the pattern, each image slice captures three distinct points where the imaging plane intersects the trident. By measuring the distances between these points ($d_l$ and $d_r$), the system can mathematically deduce the probe's pose relative to the pattern's origin ($c_{tattoo}$).
The reconstruction process involves three main steps:
- Pose Estimation: Calculating the tilt angle ($\alpha$) and the distance from the origin ($a_0$) based on the relative positions of the three detected points.
- Coordinate Transformation: Mapping the 2D image coordinates ($c_{img}$) to the global coordinate system defined by the tattoo ($c_{tattoo}$).
- Volume Compounding: Transforming all registered 2D slices into a single 3D voxel grid.
Experimental Evaluation
The method was validated using two distinct setups:
- N-wire Phantom: A box containing black nylon wires in "N" configurations submerged in a scattering medium (milk). This setup provided a ground-truth geometry to assess absolute accuracy.
- In Vivo Study: Forearm scans were conducted on three healthy volunteers to evaluate precision and reproducibility in a clinical context.
Performance was benchmarked against a high-end commercial optical tracking system (NDI Polaris Spectra).
Results and Discussion
The Tattoo Tomography method demonstrated superior performance compared to the external optical tracker. In phantom experiments, the method achieved a Fiducial Registration Error (FRE) of 0.63 mm, compared to 0.87 mm for the commercial system. In vivo results mirrored this trend, with a Mean Vessel Distance (MVD) of 0.63–0.80 mm, significantly better than the 1.06–1.40 mm achieved by the optical tracker.
A key advantage of this approach is that the tracking markers are "native" to the image modality. Because the tattoo moves with the patient's skin, the system is inherently more robust to minor patient motion than room-fixed tracking systems. Furthermore, the low cost of printed patterns makes 3D-PAT more accessible for point-of-care applications.
Limitations and Future Work
The current prototype is limited to 3 Degrees of Freedom (DoF) and assumes a flat skin surface. Future research will focus on developing more complex patterns to enable full 6-DoF tracking and incorporating surface curvature estimation to handle non-planar anatomy. Additionally, these patterns could serve as multi-modal anchors for co-registering PAT data with MRI or CT scans.
Summary updated