(C-106) Immunofluorescence-based Nuclear Morphometry Using Lamin Immunostaining and Fourier Analysis for Sensitive Detection of Nuclear Shape Abnormalities in Cancer Tissue
Introduction:: Cancer cells display morphological irregularities including changes in area, perimeter, irregularities of the nuclear periphery, thickened contours and folds. Computational extraction of nuclear morphological parameters from histological images has been explored to predict the risk of cancer progression and differentiate between lesions and normal tissue. However, the current DNA-based nuclear staining methods, such as hematoxylin-eosin (H&E) and Feulgen stain, do not adequately delineate the nuclear boundary or periphery and thus cannot efficiently capture irregularities. Here, we develop an image analysis approach for sensitive detection of nuclear shape abnormalities, based on immunofluorescence staining of nuclear lamins, which overcomes several limitations of current methods.
Materials and Methods:: Microarrays of tumor tissue were immunostained and imaged using confocal microscopy. Image analysis of nuclear morphology and marker quantification was performed using custom-developed MATLAB software. Nuclear images were first segmented by the established machine learning technique. The precise nuclear periphery tracing was accomplished by searching the pixel with the maximum intensity on each surface normal along the nuclear border. The nuclear irregularity was assessed by calculating the elliptical Fourier coefficient (EFC) ratio, which the Fourier method approximated the measured nuclear contour with a series of harmonic ellipses.
Results, Conclusions, and Discussions:: We demonstrated that the nuclear-peripheral marker staining of lamin B1, coupled with our new computational method, showed a more detailed morphology and identified irregularities of the nuclear surface more clearly than DNA staining. The number of harmonic ellipses used in the Fourier analysis was optimized by using Fréchet distance to quantify the accuracy of the Fourier analysis to reconstruct the nuclear periphery. We applied this method to the images of patient samples and cancer-adjacent tissue from various cancers including human breast adenocarcinoma, colon carcinoma, and head and neck cancer. This analysis was done specifically on epithelial cells identified in an automated fashion from cytokeratin-stained images. We observed systematic differences in EFC ratio between cells in adjacent tissue and tumor in different grades. Overall, our work demonstrates an image analysis approach for the sensitive detection and Fourier analysis of the nuclear periphery, which suggests the potential for prognostic and diagnostic applications in cancer.
