Imagens por microscopia óptica não linear para análises de biópsias de câncer canino

Abstract

According to data from the Institute of Cancer (INCA), breast cancer is a very common type of cancer among women, accounting for about 29% of cases each year. The standard diagnosis is given by visual analysis of the collected tissue, which undergoes cutting and coloring processes that can damage the sample, often making it slowly and dependent on the pathologist's experience. Thus, the development of techniques that enable a quantitative tissue analysis in order to obtain new diagnostic tools with greater agility in data analysis would be of great help to pathologists and beneficial to patients. In this work, we used the non-linear optical microscopy techniques of Second Harmonic Generation (SHG) and Two-Photon Excitation Fluorescence (TPEF) with pulsed laser to obtain images in biopsies of mammary neoplasms in female dogs. SHG images provide information of the collagen structure in the extracellular matrix which, that together with TPEF images of cell regions in the biopsies form a basis for comprehensive imaging analysis. For the image analysis, we developed a software, which uses tools already known for this type of image analysis, such as the structural tensor and the extraction of fiber networks (FIRE). The software was developed in order to automate the image analysis process. It allowed to obtain the separation of the fibrous and cellular regions in the SHG and TPEF images. With this segmentation it was possible to extract the collagen fiber networks and several specific parameters for each region (fibers or cells) forming a set of 29 metrics. With this set of metrics we were able to study the changes in the tissues caused by different types of breast cancer, such as benign mixed tumors, carcinoma in mixed tumor, ductal carcinoma In Situ, carcinosarcoma, micropapillar carcinoma and solid carcinoma in comparison to normal breast. The changes due to the tumor progression were analyzed using the statistical methods of variance (MANOVA and ANOVA). In addition, a linear discriminant analysis (LDA) including all extracted metrics allowed separating and classifying all histological types studied. We obtained a sensitivity of about 90% in the classification between healthy and cancerous tissue and a sensitivity of about 40-60% for comparison between normal breast and all the studied histological subtypes.