280 chapter 3 CT has not been investigated in thyroid nodules specifically with indeterminate cytology. In unselected nodules, some studies have been performed. Lee et al. found no significant differences between benign and malignant lesions in number of lesions, lesion size, presence of calcifications, lesion consistency, and lesion attenuation on CT in a dataset of 109 nodules (100 benign, 9 malignant) [413]. Another study in PTC found that CT was inferior to US for the evaluation of thyroid nodules [414]. More recently, AI has been investigated for CT lesion characterisation. Peng et al. investigated first order features for the identification of malignant nodules (N=50), benign nodules (N=84), and healthy controls (N=150), resulting in a sensitivity, specificity, PPV, NPV, and accuracy of 82%, 93%, 92%, 85%, 95%, and 88%, respectively [415]. It should be noted that results have not been validated using a test set. Li et al. developed a deep learning model for automatic recognition and classification of thyroid nodules on iodine contrast-enhanced CTs [416]. The model was trained in a dataset of 786 nodules (543 benign and 243 malignant) and validated in a test set of 137 nodules (103 benign and 34 malignant), resulting in an accuracy of 85%. There is a large class imbalance between benign and malignant nodules, which might have affected the accuracy, but authors state that this was corrected for using class weights. The role of CT in the preoperative differentiation of thyroid nodules is limited compared to other imaging techniques. Yet, since CT is an important source of thyroid incidentalomas (incidence: 15% [417]), computer aided detection systems to automatically recognise and classify thyroid incidentalomas on CT might be of interest. MRI Magnetic resonance imaging (MRI) is a 3D anatomical as well as functional imaging technique based on nuclear magnetic resonance [418]. MRI scanners use strong magnetic fields, magnetic field gradients, and radiofrequency waves to generate images of the organs in the body, with improved soft-tissue contrast compared to (contrast enhanced) CT. Protons (hydrogen atoms) in body tissue that contain water, give off a signal that can be processed into an image. First, a pulse of electromagnetic radiation is used to excite nuclei of atoms in the magnetic field with exactly the right resonance frequency. The excited nuclear spins of the hydrogen nucleus undergo relaxation to the ground state while emitting radiofrequency waves, which are measured with a receiving coil. The contrast between different tissues is determined by the speed at which the nuclear spin of excited nuclei returns to the ground state. Since different tissues have different hydrogen densities, details of the anatomy can then be observed. Different tissue properties can be measured using different pulse sequences of pulsed magnetic field gradients, radiofrequency pulses, intervals between delivery of successive pulses, between pulse delivery and receipt of the echo signal, etc. Intravenous contrast, mostly by paramagnetic
RkJQdWJsaXNoZXIy MTk4NDMw