Micro-tomographic and infrared spectral data mining for breast cancer diagnosis

Abstract

This study aimed to introduce a multidimensional data mining method of breast cancer (BC) diagnosis using the X-ray phase-sensitive microtomography (XPCT) and Fourier transform infrared spectroscopy (FTIR), which is a comparable approach to clinical histopathologic examination (H&E) and blood biochemical tests for achieving a simple in-situ quantitative diagnosis. Based on the high brilliance synchrotron radiation, nondestructive three-dimensional (3D) micro-tomograms of tissues and infrared absorption spectra and mapping of representative biomacromolecules of tissue and blood samples were obtained with higher signal to noise ratios (SNRs). The comparative analysis had been made between in-situ digital sections and stained histological sections. The XPCT images accurately visualized the micromorphology of tumor lesions, including fat granule degenerations, ductal proliferations, microcalcifications, collagen strands of dimensions less than 20 µm. The FTIR mapping and spectra effectively revealed the biological component distributions and tumor-related variations of lipids, proteins and nucleic acids between breast tumors and normal tissues, integrated with XPCT data into formation of complementary data. It enables to identify the complex biomolecular tumor markers, especially, the specific fine-structures of second derivative spectrum were found in 1400–1750 cm−1. Moreover, the attenuated total reflection FTIR (ATR-FTIR) data of fresh blood samples were collected to demonstrate spectral specificity and feasibility of machine learning (ML) classification for BC diagnosis compared to clinic blood biochemical tests. The random forest (RF) model was found to be the best model for differentiating BC with 100% accuracy. Therefore, the proposed methodology was proved to be a powerful tool to nondestructively visualize and classify soft tissue tumors.