Page 37 - Vibrational Spectroscopic Imaging for Biomedical Applications
P. 37

14    Cha pte r  O n e


            The classification accuracy is quantitatively measured against a
        gold standard of tissue regions selected by a trained pathologist.
        Classified images for breast TMA cores (Fig. 1.4c) are produced from
        the optimal six metric classifier obtained by iteration. Qualitative
        comparison of a classified image with the corresponding H&E image
        from an adjacent TMA cut in Fig. 1.4a indicates a reasonable correla-
        tion between classified images and H&E-stained images. These clas-
        sified images and ROC analysis results indicate that FT-IR imaging
        coupled with Bayesian classification has the potential to reliably
        select tissue classes that correspond with H&E staining. This high-
        throughput spectral classification approach that incorporates mil-
        lions of spectra from many patients is the first step to establish spectral
        image classification as a diagnostic tool in a clinical setting.

        1.2.3  Spectral Metrics and Biochemical Basis
        The six metrics selected for stroma and epithelium classification are
        displayed in Table 1.1. These spectral features can be used as an opti-
        mal classifier with an AUC of 0.995 for calibration data. Nearly all of
        these metrics fall in the spectral “fingerprint region.” This region con-
        tains many narrow overlapping peaks, and is often useful for
        identifying complex molecules. Therefore, it is not surprising that



                                                        Molecular
                          -1
    Feature     Position (cm )  Assignment 33           Origin
    Peak ratio  1080:1456     1080 cm : symmetric       DNA/RNA
                                      −1
                              PO  stretching, CO stretching   Protein
                                 –
                                 2
                                      −1
                              1456 cm : assymetric CH
                                                   3
                              bending
                                      −1
    Peak ratio  1556:1652     1556 cm : NH bending,     Protein
                                                 −1
                              CN stretching 1652 cm :   (Amide I &
                              CO stretching             Amide II)
                                     −1
    Peak ratio  1080:1238     1080 cm : symmetric PO  –  DNA/RNA
                                                  2
                              stretching, CO stretching 1238
                                 –1
                                             –
                              cm : assymetric PO  stretching
                                             2
                                      −1
    Center of   1216–1274     1236 cm : NH bending,     DNA/RNA
    gravity                   CN stretching, CH  wagging,   Protein
                                            2
                                          –
                              assymetric PO  stretching  (Amide III)
                                          2
                                      −1
    Peak ratio  1338:1080     1080 cm : symmetric       DNA/RNA
                                 –
                              PO  stretching, CO stretching  Protein
                                2
                                      −1
                              1338 cm : CH  wagging     (Amide III)
                                          2
                                      –1
    Peak area   1426–1482     1456 cm : assymetric CH   Protein
                                                   3
                              bending
   TABLE 1.1  Spectral Metrics Selected by Optimization of the Stroma and
   Epithelium Histology Classification Model
   32   33   34   35   36   37   38   39   40   41   42