Physics-Based Modeling and Inverse Problems in Optical Imaging
Y. Liu
École polytechnique fédérale de Lausanne, EPFL Thesis no. 10801 (2026), 190 p., March 13, 2026.
At the intersection of computational imaging and biomedical optics, this thesis applies physics-based modeling and computational methods to solve inverse problems in optical imaging. We focus on three representative imaging modalities where we push the limits of imaging depth and spatial resolution: optical projection tomography (OPT); photoacoustic imaging (PAI); and single-molecule localization microscopy (SMLM). OPT and PAI overcome the diffusion limit to enable greater imaging depths, while SMLM breaks the diffraction limit to achieve super-resolution.
The first modality, OPT, produces high-resolution 3D images of samples ranging from millimeter to centimeters, but the quality of images is sensitive to mismatches between the model and experimental setup. We propose refined mathematical models, along with computationally efficient calibration and reconstruction algorithms. This work helps OPT practitioners to improve the resolution of their images by identifying the root of their artifacts and correcting them.
The second modality, PAI, combines optical excitation and acoustic detection. It relies on the labeling of photo-switching protein reporters to achieve high-specificity lock-in detection of these labels, at millimeter depth. We propose a detailed mathematical model of the evolution of the detected signals. It allows us to develop an unmixing algorithm based on sparse regularization to retrieve concentration maps of the proteins directly from acoustic measurements.
Finally, SMLM enables localization of molecules at nanometer resolution. There, careful design and accurate modeling of the point spread function (PSF) are important to many applications. We first present a fast and low-cost optimization pipeline to investigate illumination profiles that achieve the best localization precision for MINFLUX, a recent SMLM technique based on carefully engineered excitation patterns. Then, we propose a unifying framework for high-numerical-aperture vectorial PSF models and an efficient implementation as an open-source library, PSF Generator.
In summary, this thesis demonstrates that rigorous models, tailored algorithms, and efficient implementation are fundamental for the improvement of the image quality in optical imaging techniques.
@PHDTHESIS(http://bigwww.epfl.ch/publications/liu2601.html,
AUTHOR="Liu, Y.",
TITLE="Physics-Based Modeling and Inverse Problems in Optical Imaging",
SCHOOL="{\'{E}}cole polytechnique f{\'{e}}d{\'{e}}rale de {L}ausanne
({EPFL})",
YEAR="2026",
type="{EPFL} Thesis no.\ 10801 (2026), 190 p.",
address="",
month="March 13,",
note="")