3D OPTO-ACOUSTIC IMAGE RECONSTRUCTION AND MOTION TRACKING USING CONVEX OPTIMIZATION ALGORITHMS

Supervisor: Dr. Michael Kolios

This work involves developing techniques for improved opto-acoustic imaging (OA), with the goal of enabling enhanced visualization of vascular structures in biological tissue. This has potential application for diagnosis and detection of cancer and other diseases where blood vessels have structural and functional differences from healthy tissue. In OA systems, acoustic waves are generated by absorption of optical energy. Since hemoglobin absorbs more light than other molecules in tissue, images of the tissue’s blood distribution can be reconstructed by processing measured acoustic signals. Moreover, the wavelength-specific optical absorption of oxy- and deoxy-hemoglobin permits OA to image the blood’s oxygen saturation level. However, OA image quality is limited by the ability to localize acoustic sources in tissue, and by the ability to collect sufficient data to accurately reconstruct tissue properties. To improve OA image quality, this work investigates using convex mathematical optimization to perform image reconstruction from transducer measurements. The proposed technique iteratively solves an inverse problem by fitting the measured data onto simulated OA signals. To accelerate computational performance, mathematical simplifications for 3D simulation and reconstruction are developed. Using multiple acquisitions to provide 3D volumetric information, a method is developed to determine transducer motion from OA data during image reconstruction. In addition, the ability to visualize blood oxygen saturation is characterized for a clinical OA breast imaging device, and image quality is studied using biologically-relevant tissue phantoms. Results demonstrate that reconstruction with mathematical optimization can achieve higher contrast-to-background ratio (CBR) and peak-signal-to-noise ratio (PSNR) compared to approaches involving backprojection. In addition, using a separable model for the system’s response reduces computational complexity by a factor of n in a 3D volume with n³ voxels. This potentially enables faster and more accurate image reconstruction in OA systems.