Diagnostic Analysis and Modeling of Trace Biometals and Speciation Alterations in Urinary Tract Cancers by Machine Learning Based X-ray Fluorescence and Scatter (Xrfs) Spectrometry

Abstract

X-ray fluorescence spectroscopy has the capability to determine the levels, chemical speciation, and distribution of trace biometals in a biological sample for disease diagnostics. Direct biometals analysis in soft tissues and fluids by X-ray fluorescence for early cancer diagnosis has hardly been explored due to dark matrix challenges that results to weak analyte signals as well as intricate multivariate relationships. This study was aimed to develop a machine learning-aided X-ray fluorescence and scatter method for early diagnosis of urinary tract cancer (prostate and urinary bladder) based on the concentration, speciation and 2D imaging of trace biometals in cell culturs, human tissue and urine. The XRF variants enabled simultaneous analysis of biometals’ levels, speciation and 2D distribution in; simulate tissue and urine, human tissue and urine, and cultured samples. The levels of the biometals (Mn, Cr, Cu, Fe, Zn and Se) were determined by multivariate calibration model (ANN) using EDXRF selected fluorescence and scatter regions from simulate urine and tissues. The levels of Fe, Cu and Zn in human tissue biopsies were in the range of; 150.0±4.5-181.2±9.5 ppm, 16.4±5.4-25.9±2.6 ppm and 60.5±12.4-90.2±3.8 ppm respectively with alteration in levels of Cr, Mn and Se. High (p<0.001) concentrations of Fe, Cu and Zn were noted in cultured cancer tissues compared to normal human tissues and urine. The KNN distinguished the chemical speciation of Cu and Fe in cancerous and normal urine at 90% classification accuracy. The human cancerous urine samples were found to be rich in Fe and Cu occurring mostly as 2Fe and Cu possibly due to their oxidative role in Fenton reactions that accelerates carcinogenesis. SR-μXRF enabled 2D spatial distribution of trace biometals (Mn, Fe, Cu and Se) patterns where significant (p<0.05) differences in the accumulation of the analytes in cancerous and healthy cell cultures. The 2-D spatial distribution maps of the trace biometals revealed high spatial correlation between Cu and Fe (0.9406) and (0.92252) in DU_D3 and DU_D4 respectively in cancerous compared to corresponding stages in normal cell cultures for cancer diagnosis. Artificial Neural networks (ANN) distinctively classified cell cultures into cancerous and healthy groups by PC1 and PC2 scores of Cu and Fe L-lines spatial-spectral profiles. Further utility of selected fluorescence and scatter spectral profiles enabled the classification of cancerous cultured cells into early, intermediate and advanced stages of cancer development. The study demonstrated the utility of machine learning-aided XRFS analysis of hardly discernible fluorescence peaks and Compton scatter in tissues, cell cultures and urine samples to realize rapid non-invasive cancer diagnostic model.