Date of Award
12-22-2023
Degree Name
Doctor of Philosophy (PhD)
Degree Type
Dissertation
Abstract
Disease states are the result of a complex interplay of many different cell types interacting in close proximity in the context of often heterogeneous tissues. Alpha particles are drawing intense research and clinical interest because of their potent cytotoxic effects and their short path lengths. Analyzing the dose distribution and tissue micro-environment for alpha therapy plays a key role in predicting the efficacy of this targeted radiotherapy. However, to date there have been no direct on-tissue analytical methods for alpha dose distributions and the corresponding tissue microenvironments. Therefore, we have developed a pipeline to overcome this limitation by utilizing quantitative imaging techniques including digital autoradiography (DAR) and imaging mass cytometry (IMC). DAR is an imaging technique that provides information on the distribution of radionuclides in tissue sections and is widely used in drug development. IMC is an emerging highly multiplexed molecularly specific histological method that can report the protein expression profiles of up to 40 markers on processed tissue sections. DAR and IMC have their own deficiencies that have challenged their utility in previous applications. Firstly, DAR suffers from low image resolution and significant background noise, which can lead to poor correlation and, in some cases, errors in determining the relationship between radiotracer distribution, anatomical structure, and molecular expression profiles. Secondly, despite extensive optimization of staining conditions, IMC images may exhibit low signal-to-noise ratios (SNR) for specific markers, and the presence of pixel intensity artifacts can detrimentally affect image quality and subsequent downstream analysis. Lastly, the spatial information of cells obtained from IMC images has not been adequately harnessed, resulting in a limited understanding of higher-order patterns such as tissue organization. To enable direct on-tissue analysis of dose distribution and the tissue microenvironment, we propose the development of algorithms and pipelines to address these challenges. First, we introduce a Poisson-Gaussian penalized expectation maximization (PG-PEM) algorithm to blindly enhance DAR images, thereby improving dose measurement accuracy. Subsequently, we present IMC-Denoise, a content-aware pipeline designed to enhance IMC cell phenotyping outcomes. It includes a differential intensity-based restoration algorithm (DIMR) for outlier pixel removal and a self-supervised deep learning algorithm for shot noise image filtering (DeepSNiF). Furthermore, we introduce an interpretable spatial cell learning framework called InterSTELLAR, which classifies tissues into distinct clinical types. InterSTELLAR incorporates an attention-based pooling module for cell-level interpretable learning. Finally, with the improved data and newly developed algorithms, we establish an integrated pipeline for the automated analysis of data from different modalities. These innovative approaches are poised to enhance our understanding of dose distribution and tissue microenvironment responses, benefiting target engagement studies in drug development and enabling more precise theranostic medicine.
Language
English (en)
Chair
Daniel Thorek