| dc.description.abstract | Photoacoustic imaging (PAI) employed the special properties of light or photons to obtain
detailed images of organs, tissues, cells, and even molecules. The method allowed for a
non-invasive or minimally invasive examination within the body. The PAI used nearinfrared
light (600 nm - 900 nm) as the scan media, which had the additional benefit of
being a non-ionizing imaging modality. The PAI can be integrated with other imaging
modalities, such as MRI or X-ray, to provide better information for complex diseases or
researchers working on complex experiments. Photoacoustic imaging has already been
widely used in pre-clinical research to image small animals. Although PAI is a multi-scale
modality, it is challenging to use for clinical research and interventional applications due
to the non-linear distribution of optical fluence.
Quantitative Photoacoustic Imaging (QPAI) has remained problematic due to the
influence of non-linear optical fluence distribution, which influences photoacoustic image
representation. Non-linear optical fluence correction in PA imaging was highly ill-posed,
leading to inaccurate recovery of optical absorption maps. Note that the traditional
optical fluence correction method needs precise estimation of optical fluence map. Many
different light transport models exist for estimating the optical fluence map when the
optical properties, i.e., optical absorption and optical scattering are known. However,
in reality the optical properties are unknown in advance, therefore fluence estimation
becomes difficult during PA imaging. Moreover, optical light illumination at the target
medium under the study is not uniform over the wavelength, the target medium introduce
spectral distortion between the measured PA spectrum and the true target spectrum.
Consequently, for true QPAI, the optical fluence must be simultaneously estimated and
compensated. This requires not only an appropriate fluence model, but also an effective
method to estimate the fluence distribution at each wavelength from PA measurements.
Based on prior knowledge of the target medium’s optical properties, many different
methods have been proposed for fluence compensation for a simple and homogeneous
medium. Unfortunately, none translate into clinical usage. To translate to clinical usage,
more complex and heterogeneous media need to be studied. And also the generated PA
signal may also change dynamically based on the background tissue properties. Hence,
consider complex, foreground and background non-homogeneity of the medium for accurate
recovery of optical absorption coefficient. None of the research groups adapted all
the above factors simultaneously for fluence compensation. This thesis study developed
a deep learning-based optical fluence correction approach to solving this limitation.
The main objective of this thesis was to investigate the non-linear distribution of optical
fluence effect in 2D and 3D medium and compensate this effect by using deep learning
(DL) models. This thesis explains the recovery of the optical absorption maps using
deep learning approaches by correcting the fluence effect. In this thesis, different deep
learning models were compared and investigated to enable optical absorption coefficient
recovery at a particular wavelength in a non-homogeneous foreground and background
medium. Data-driven models were trained with two-dimensional (2D) Blood vessel and
three-dimensional (3D) numerical breast phantom with highly heterogeneous/realistic
structures to correct for the non-linear optical fluence distribution. The trained deep
learning models like U-Net, FD U-Net, Y-Net, FD Y-Net, Deep ResUnet, and GAN
were tested to evaluate the performance of optical absorption coefficient recovery with
in-silico and in-vivo dataset. The results indicated that DL-based deconvolution improves
the reconstructed PAI in terms of PSNR and SSIM. Further, it was observed
that DL models can indeed highlight deep-seated structures with higher contrast due
to fluence compensation. Importantly, the DL models were found to be about 17 times
faster than solving diffusion equation for fluence correction and also able to compensate
for nonlinear optical fluence distribution more effectively and improve the photoacoustic
image quality. | en_US |
