Thesis of Ruggeor Gurrini

Defense of thesis Ruggeor Gurrini - laboratory LASIRe

Abstract : 

Today, hyperspectral imaging (HSI) plays a key role among the analytical techniques. The challenge is to improve both spatial and spectral resolution as well as acquisition speed. In this situation, chemometrics plays a key role as a relevant tool to extract meaningful information from the data. MALDI and LIBS can be considered the most used in molecular and elemental imaging respectively. Although they share the same problem of data dimension, the different data acquisition process, the different type of data, data storage, as well as the different purpose of analysis explain the need for different tools between LIBS and MALDI. In MALDI, as a non-target analysis, clustering has more quickly become one of the most important techniques studied due to its ability to group similar objects, e.g. similar region in a MALDI image. In diagnosis, tissues are usually examined by the golden standard of histopathological and immunohistochemical analysis. However, these are targeted analyses, so they are not suitable if we are looking for new biomarkers. In fact, MALDI can provide complete spatially resolved biological information. With this amount of information it is possible to segment the image to a region with similar biochemical information. A common technique it is to follow the histological annotation to cluster the data with an interactive hierarchical clustering technique (bisecting kmeans), or even if there is no histological annotation the segmentation is mainly driven by the user experience, so it can be really biased. For this reason, one of the main purposes of this research was to find an appropriate chemometric approach to cluster the MALDI data in a less biased way, using a hierarchical approach, giving all the sheets to be studied and evaluating their division without using prior information such as histological annotation to match or premonitions. LIBS is widely used in various fields. Millions of spectra can be measured every day, which is a real challenge for data set analysis. In the field of mineral analysis, a sample can contain multiple phases and different minerals can have very close elemental compositions, making data analysis really difficult. Even if the composition does not appear to be different, they can appear to be different from a visual perspective. This discrepancy is important to investigate because it can help to really understand and explore the sample. For this reason, the HSI data fusion approach with LIBS and RGB has been explored to enhance the LIBS analysis. In addition, the possibility of fusing elemental and molecular information in a biological context is being explored. As the most used in its field, the fusion between MALDI and LIBS can enhance the exploration of biological tissue, allowing elemental and molecular information to be correlated, with the possibility of understanding whether exogenous elemental modifications can cause molecular variations related to disease. As this is a pioneering study, we started with a rat sagittal brain sample to see the feasibility of the technique. Given the state of the art, the work was not so much focused on developing brand new advanced methods, but rather on changing the approach and pose the right research question.

Keywords : Hyperspectral imaging, Clustering, Chemometrics, MALDI, LIBS, machine learning