Embedded Files
Photoacoustic Tomography: Listening to The Sound of Tissues
Photoacoustic Tomography: Listening to The Sound of Tissues
Photoacoustic tomography (PAT) is a rapidly growing biomedical imaging modality. In optical imaging technologies, light scattering limits the spatial resolution in deep tissues. PAT combines the benefits of both optical and ultrasonic imaging to produce high-resolution, high-contrast images of deep tissues. The imaging process of PAT starts with illumination of a biological tissue with short-pulsed laser. As light propagates into the tissue, it is absorbed by its biomolecules. This absorbed optical energy is converted into heat and in the result of thermoelastic expansion of the tissue, pressure wave starts to propagate as an ultrasound (acoustic) signal. This acoustic signal can then be detected by ultrasonic transducers outside the tissue to make an image from desired tissue.
Imaging principle of PAT. Laser pulses illuminate the target tissue and the resultant ultrasound signal is detected by a transducer.
PAT can be used for functional, structural and molecular imaging from organelles, cells, tissues, to organs. A variety of practical applications have been explored to show PAT great potential in preclinical and clinical imaging, such as small animals whole-body imaging, breast cancer detection, tumor angiogenesis monitoring, brain imaging, skin cancer detection, etc.
Image reconstruction is one of the basic foundations of each medical imaging technology including PAT. However, in many practical applications, geometrical limitations during photoacoustic signal acquisition or in order to accelerating this process causes unwanted artifacts, which leads to low-quality images.
Applying standard algorithms to sparse data PAT that yields low-quality images with limited-view artifacts.
The rising interest in deep learning techniques for image reconstruction has led to a transition from classic reconstruction methods to data-driven and learning-based approaches. These learning-based methods have been successful in enhancing PAT reconstructed images, too. Convolutional neural network especially U-net architecture has been widely used for PAT image reconstruction and has shown great success in artifact removing and improving image quality. Furthermore, Generative adversarial networks (GANs) are also powerful technology within the space of deep learning with many great applications in particular for generating new plausible images. In our work for PAT image reconstruction from sparse data, by considering image reconstruction as a kind of image-to-image translation task, that is from an artefactual image to an artifact-free image., we have used GANs to reach better reconstructions in terms of image quality.
Page updated
Report abuse