278 chapter 3 for some patients, the use of radiomics could possibly circumvent the need for FNAC and surgical resection [407]. For deep learning radiomics using convolutional neural networks (CNN), the pooled sensitivity and specificity were 85% and 82%, significantly lower than for studies using non-CNN (sensitivity: 90%, specificity: 88%), which might be due to a larger required sample size for a deep learning radiomic study (at least 800) compared to a handcrafted radiomic study (around 100) [392]. The meta-analysis only touches upon the heterogeneous methodology of included studies, stating the broad spectrum of analysis methods and interobserver variability of US. Notwithstanding, radiomic features extracted from US images are impacted by the slice variability and pre-processing [408]. To improve feature repeatability, the use of intensity standardisation with outlier removal applied to the region of interest and a fixed bin size grey-level discretisation could be performed and these and other pre-processing steps should be extensively documented [408]. When standardisation of the radiomic methodology is performed and US radiomics is validated in large prospective cohorts, it has the potential to become a non-invasive and cost-effective diagnostic tool in (cytologically indeterminate) thyroid nodules. Elastosonography One of the key features during palpation of thyroid nodules is the degree of firmness; malignant nodules tend to be firmer than benign ones. Palpation, however, is highly subjective and depends on the size and location of the nodule and on the skill of the practitioner. Elastosonography, a dynamic US technique that is used to evaluate the biomechanical viscoelastic properties of tissue, provides a quantitative method to measure tissue firmness or elasticity. Lyshchik et al. were the first to practice elastosonography for the evaluation of the elasticity of thyroid nodules, measuring the tissue distortion while applying a standardised dosed external force by the US transducer [281]. Elastosonography methodology is diverse, but it follows the principle of estimating displacements fields in tissue using correlation techniques that track the echo delays in waveforms recorded before and after the quasistatic compression. Qualitative evaluation of the thyroid elasticity is performed by repeated manual compression (also known as strain elastosonography), taking into account the amount of compression and different zones of interest (i.e., healthy tissue should be included in the measurement, which might be complicated in the presence of thyroid diseases or large nodules) [409]. Alternatively, and circumventing the problem that the mechanical compression force applied to the tissue cannot be measured accurately and thus the absolute tissue strain cannot be calculated, shear wave elastosonography has been developed. This technique evaluates tissue stiffness through focused pulses of US instead of mechanical compression [410]. This acoustic force causes horizontal displacements in the tissue, which are called shear waves. These shear waves contain quantitative data about the elastic properties of the tissue that can be measured in propagation speeds of these sheer wave (m/s) or nodule stiffness (kilopascals). It has the advantages of being more objective, having a higher reproducibility, and having decreased operator dependence.
RkJQdWJsaXNoZXIy MTk4NDMw