Chapter 10 this, feature Ft36 was established. Another blood‐related feature (Ft3) in the Si sensor range was complementary for distinguishing parathyroid from surrounding adipose tissue (Figure 10.4). Within the InGaAs sensor range fat‐related features (Figure 10.5, respectively Ft12 and Ft14, and Ft12 and Ft17) performed best for enabling automated tissue classification. Combining Si and InGaAs sensor features resulted in improved tissue classification, this was especially profound when discriminating parathyroid from adipose tissue. The statistically different spectral features used for classification of parathyroid versus adipose tissue or thyroid are probably due to differences in chemical composition (e.g., water/lipid content) of the investigated tissues and structures. Experimental methods such as near‐infrared fluorescence imaging after peripheral infusion of aminolevulinic acid8,9 or methylene blue10 (i.e. exogenous contrast agents) can be used for parathyroid gland or parathyroid adenoma localization. The reflectance spectra in our study, which were the basis to extract spectral features, originate from intrinsic tissue properties (i.e. endogenous chromophores) that do not require dye administration. Consequently there is no risk of potential toxicity or allergy to a contrast agent. As the LOO cross‐validation method inherently produces relatively optimistic classification results, external validation remains essential before classification models can be implemented in clinical practice37. Such validation would need to be performed on newly acquired data in a multi‐center study. In this study, we use physiological knowledge on composition of different tissue types, and thereby use the pre‐defined spectral regions of interest covering hemoglobin, water and fat. Tissue differentiation on the basis of completely automatically extracted features from a larger data set might achieve better results, and should be explored in future work. Additionally, as the acquired reflectance spectra are specific to the probe geometry, more research is needed on the relation between these reflectance spectra and in vivo intrinsic tissue biological properties which potentially could be captured by optical tissue properties. This is needed to obtain a better understanding of the nature of discrimination performance: i.e. whether parathyroid glands can be optically distinguished from surrounding adipose tissues or adjacent thyroid tissue based on differences in light scattering behaviour (related to structural differences) or due to differences in absorption behaviour (related to chromophore concentrations such as blood, water and fat). Such an approach would also be tissue‐specific and robust to inter‐patient and multi‐center variability. The gold standard used in this study was the surgeon’s visual judgment (one surgeon who has performed >1000 thyroid surgeries; histo‐pathological confirmation of the in vivo measured tissues was not possible). This judgment is based on color (spectral) information, on the recognition of anatomical position as well as on spatial structure of 160
proefschrift_Schols_SLV
To see the actual publication please follow the link above