270 chapter 3 CT also allows for quantification, as attenuation coefficients of tissues are linearly transformed to Hounsfield units (HU), where a value of 0 represents the attenuation coefficient of distilled water and a value of -1000 represents the attenuation coefficient of air. However, in practice, deviations in this linearity occur. Increasing the tube voltage and with that the photon energy generally decreases the probability of interactions, i.e., attenuation and, therefore, increases penetration. Also, different scanners deliver different tube currents or photons to the subjects for a given milliamperage × seconds (mAs), as a consequence of differences in beam filtration, variances in tube potential, and rotation times [364]. Consequently, a fixed milliamperage yields different exposures, resulting in noise differences and inconsistencies in HU measurements. Other critical factors include spatial and temporal resolution, reconstruction kernel, subject positioning within the CT scanner bore, breath-holding techniques, and the (frequency of) monitoring of the CT scanner calibrations (i.e., quality control procedures). No central accreditation programmes have been ventured yet, but harmonisation has been attempted in specific applications [364]. Quantitative analysis of MRI is even more complex, due to the relative scale of the so-called weighted images. Image contrast is affected by factors intrinsic to the tissue, specific to the examination, and dependent on the hardware. Also, conventional MRI techniques lack biological specificity, i.e., different physiological and pathological substrates can produce similar changes in image contrast. MRI studies can be quantified by obtaining parametric maps of meaningful physical or chemical variables (e.g., apparent diffusion coefficient, ADC) that can be measured in physical units (mm2/s for ADC) and compared between tissue regions and among subjects. Like for CT, only local initiatives aim to harmonise images [365, 366]. Conventional US is qualitative in nature, but quantitative US can provide specific numbers related to tissue features that can increase the specificity of image findings [367]. Qualitative bright mode (B-mode) US displays a morphological representation of the tissue, obtained from the radiofrequency data. Quantitative US, on the other hand, processes the raw radiofrequency data from tissue backscatters to characterise and distinguish phenotypic changes at a cellular level. Other US techniques like spectral-based parameterisation, elastography, shear wave imaging, flow estimation, and envelope statistics can also be performed quantitatively. However, most clinical devices do not incorporate quantitative US yet. Artificial intelligence Recent developments in computer science have led to advanced artificial intelligence (AI) approaches, capable of capturing the information concealed in the image in the interest of lesion or disease detection, classification and diagnosis, segmentation, image reconstruction and quantification [368]. An important breakthrough within AI was the advancement of machine
RkJQdWJsaXNoZXIy MTk4NDMw