In a collaborative study between the Indian Institute of Science (IISc), Bangalore, and the University of Twente, The Netherlands, researchers have designed a new algorithm for image recovery in Photoacoustic Tomography (PAT). PAT is an important non-invasive biomedical imaging technique where the optical contrast rendered by laser beams and the superior resolution of ultrasound waves are used to study biological tissues. The new algorithm works better with higher accuracy as compared to the conventional ones in use today.
Non-invasive imaging is important in biological research and in the diagnosis and treatment of various diseases. In imaging using PAT, a laser light is first shone on the target tissue. Oxygenated haemoglobin, non-oxygenated haemoglobin, water and lipids are the major light absorbing agents (chromophores) in living tissue. As light energy is absorbed, the target tissue undergoes a temporary elastic expansion. This leads to the emission of acoustic waves due to a phenomenon called the ‘photoacoustic effect’. The emitted ultrasonic waves are then detected by ultrasonic transducers placed outside the tissue and are further processed to generate a complete image. This technique provides high-resolution structural, functional, and molecular imaging of organelles, cells, tissues, and organs. It also provides details on the physiological properties, such as the blood haemoglobin concentration and oxygen saturation.
Image reconstructions, in techniques such as PAT, rely on the powerful algorithms that work in the background to conjure up the image from the measured data –the acoustic waves in this scenario. Since the equation governing photoacoustic wave propagation and the relation connecting the initial acoustic profile to the optical absorption distribution are both known, a reconstruction algorithm should aim to directly estimate this absorption map using the measured acoustic signals. Nevertheless, most of the existing algorithms either recover only the initial acoustic profile or rely on the reconstructed acoustic profile to subsequently estimate the optical absorption map in a second step. This two-step approach can result only in qualitative images due to the undesirable dependence of the initial acoustic profile on the optical source positions.
The researchers have now designed a one step approach to solve these challenges. “Most of the work in PAT is focused on developing better instruments for imaging and there is less stress on the development of better algorithms”, points out Ms. Mamatha Venugopal, a research scholar from IISc who is a member of the study group. Though one-step recovery of the optical absorption map is believed to be advantageous, such algorithms have not been attempted due to their hectic computational load. Ms. Venugopal and her team have tackled this problem by proposing a 'stochastic search' algorithm for the one step recovery of the optical absorption map from the measured acoustic signals. This method not only delivers quantitative images with an improved accuracy, but also side steps the feared computational load.
For the reconstruction, the area to be imaged is first broken up into a grid and each of the boxes is assigned a variable as its absorption parameter. A vector is constructed with these variables as its components, which defines the overall absorption profile. In the newly developed algorithm, this absorption vector is modelled as a stochastic process evolving with time. In the first step of the algorithm, the absorption profile vector is assigned a random distribution, say one in which all grids have zero absorption. The acoustic wave profile can be calculated by solving the photoacoustic equations with the absorption profile as the input. In the subsequent steps, the absorption profile vector is iteratively modified such that, with every iteration, the misfit between the measured and calculated acoustic profiles approaches zero. The researchers found that the quantitative accuracy of the recovered absorption map using the new algorithm from both numerical and experimental data is good with an overall error of less than 10%.
As a next step, the researchers aim to develop such algorithms for the direct recovery of exact chromophore concentrations from the acoustic data. Further, plans to progress on to animal imaging and clinical data are under way too. “In the long term, we want to look at the vascular abnormalities and blood oxygenation levels in the fingers of patients with rheumatoid arthritis, using PAT”, says Ms. Venugopal. Armed with a better algorithm, the researchers expect to shed some light on the exact causes of inflammation in arthritis patients, which remains a topic of contention in the biomedical community.