Multimodal Biophotonics Imaging of Cancer Biomarkers

Background: Early stages of cancer are typically characterized by changes in morphology and metabolism on a cellular level and remain largely undetected by current non-invasive screening techniques. Optical methods, such as Multiphoton Microscopy (MPM), have the ability to detect these changes non-invasively.

In order to develop diagnostic tools, such as endoscopes, measurable parameters need to be defined as biomarkers for specific diseases. Moreover, an accessible test environment is needed to evaluate these tools and quantify their diagnostic accuracy. Aim: To establish biomarkers and provide a test environment for said diagnostic devices, three core hypotheses are investigated in this thesis: (1) Variations in autofluorescence spectra can be used to create a fluorescence ratio and serve as a functional biomarker. (2) Spectral bands can be defined for a specific application, which allows for optimal cell and tissue classification using a simplified detection. (3) Structural features, in combination with functional biomarkers, can improve classification and allow for a quantitative comparison of diagnostic tools.

Approach: Multicellular spheroids of cancerous (HT29) and noncancerous (FHC, HCoEpiC) human epithelial colon cells are used as tissue models. Two MPM systems are used for biomarker characterization. A custom build hyperspectral microscope with a fixed excitation wavelength at 785 nm is developed based on an existing setup by adding monochromatic imaging, modifying software for automated acquisitions and improving signal filtering as well as excitation and collection optics.

A second MPM system with tunable excitation from 750 nm to 1100 nm is modified to extract autofluorescence spectra from selected tissue regions. Tools for sample handling are developed and an algorithm is written to determine optimal spectral detection bands and their emission wavelengths for tissue classification.

Structural biomarkers are investigated on TPEF images by evaluating the power spectral density to estimate a nuclear density and quantify mitochondrial clustering. Finally, biomarkers are combined in a multivariate analysis to quantify diagnostic accuracy Results: This thesis confirms the aforementioned hypotheses: (1) Differences in autofluorescence spectra allow for a separation of cancerous and normal tissue models. (2) The use of optimized spectral bands at fixed excitation improves separation by 31.5% while accepting a reduced SNR in the images.

Moreover, by tuning the excitation wavelength, an additional 2.5 fold increase in separation is predicted, while the SNR of the image remains constant. (3) Combing an analysis of the fluorescence ratio with information about nuclear density and mitochondrial clustering improves the combined tissue classification by a factor of 2.

Furthermore, the complementary nature of these three biomarkers enables a quantitative comparison of classification accuracy depending on image modalities and instrumentation parameters. Conclusions: The established framework allows for a definition of disease-specific biomarkers and an evaluation of diagnostic tools.

Additionally, an improved classification performance was shown in individually optimized biomarkers and combinations thereof. The methods presented in this work can be used as a tool to establish correlations between imaging modalities and to benchmark the classification accuracy on an accessible tissue model.