FTIR Spectroscopic Imaging Supports Urine Cytology for Classification of Low- and High-Grade Bladder Carcinoma

Abstract

Bladder urothelial carcinoma (BC) is a common, recurrent, life-threatening, and unpredictable disease which is difficult to diagnose. These features make it one of the costliest malignancies. Although many possible diagnostic methods are available, molecular heterogeneity and difficulties in cytological or histological examination induce an urgent need to improve diagnostic techniques. Herein, we applied Fourier transform infrared spectroscopy in imaging mode (FTIR) to investigate patients' cytology samples assigned to normal (N), low-grade (LG) and high-grade (HG) BC. With unsupervised hierarchical cluster analysis (UHCA) and hematoxylin-eosin (HE) staining, we observed a correlation between N cell types and morphology. High-glycogen superficial (umbrella) and low-glycogen piriform urothelial cells, both with normal morphology, were observed. Based on the spectra derived from UHCA, principal component analysis (PCA) and partial least squares discriminant analysis (PLS-DA) were performed, indicating a variation of protein content between the patient groups. Moreover, BC spectral cytology identified a low number of high-glycogen cells for which a shift of the carbohydrate/phosphate bands was also observed. Despite high cellular heterogeneity, PLS-DA was able to classify the spectra obtained. The voided urine FTIR cytology is one of the options that might be helpful in BC diagnosis, as high sensitivity and specificity up to 97% were determined.

Keywords: bladder carcinoma; cytology; diagnostics; infrared spectroscopic imaging.

Figure 1

Comparison of the HE microphotographs of normal, LG BC and HG BC cytology (magnification: 100× for normal and LG BC samples and 50× for HG BC) with false-color UHCA maps from IR images and their mean absorbance spectra. The colors of the spectra correspond to the colors of UHCA classes.

Figure 2

(A). Averaged second derivative FTIR spectra (±SD—dashed lines) and PC 1-3 loadings plots from principal component analysis (PCA). (B). 3-dimensional PCA scores plot. Averaged spectra and PCA were calculated from 270 spectra for each patient group (810 in total).

Figure 3

The integration of: (A) glycogen (1153 cm⁻¹), (B) proteins (1652 cm⁻¹), and (C) carbohydrate/protein ratio. Integral intensities were calculated from second-derivative FTIR spectra; 270 spectra per each group (810 in total).

Figure 4

(A) PLS-DA loadings (N with positive scores). (B) Predicted values, calculated in pairs of N vs. LG BC and N vs. HG BC (red—SD, blue—unexplained by model variance). A total of 135 derivative FTIR spectra from each group (9 spectra from each patient) were taken to build the model, and the remaining 135 spectra were used for prediction. The spectra above zero are predicted as N, and spectra below zero are predicted to be LG or HG BC, respectively.

Figure 5

Confusion matrices for N vs. LG BC groups and N vs. HG BC, calculated from PLS-DA results. Abbreviations: TP—true positive, FN—false negative, FP—false positive, TN—true negative, NegPredVal—negative predictive value.