Thyroid nodule ultrasound: technical advances and future horizons

Thyroid nodule detection, margin delineation and size measurement are heavily influenced by spatial resolution. Modern high-frequency US probes utilise piezoelectric material that resonate at higher frequencies and across wider bandwidths (i.e., PureWave^TM, Philips) than older crystals. Improving the source piezoelectric material enables the effective bandwidth of the probe to be used more efficiently and uniformly than was previously possible [3]. The resulting improvements in spatial resolution are particularly beneficial in superficial structures such as the thyroid. In addition, probes now contain higher numbers of elements (e.g., high-density 18 L6, Siemens ACUSON) facilitating more versatile beam forming and providing the basis for increasingly advanced compound imaging. Spatial compounding refers to composite images created by insonations from transmitted pulses at multiple different angles to the probe surface (Fig. 1) whilst frequency compounding generates images based on a range of frequencies across the probe bandwidth [4]. The two techniques are often used simultaneously, producing higher resolution images with smooth tissue planes and increased signal-to-noise ratios. Traditionally, these methods of improving spatial resolution have come at a cost: a reduced frame rate (temporal resolution) due to the conventional ultrasound transmit-receive cycle. The newest technical developments aim to minimise this limitation; an example is the use of precision beam-forming, multi-line transmission and parallel processing (nSIGHT^TM, Philips) to combine high spatial and temporal resolution. As a final consideration, the availability of high-definition, multi-position monitors enables the information acquired and processed by the transducer to be clearly displayed for interpretation.

Contrast resolution settings are the primary factors that influence the assessment of nodule consistency, reflectivity and the detection of echogenic foci due to calcium or colloid. Whilst broader bandwidths allow improved beam penetration in the neck, the majority of advances in this aspect of thyroid ultrasound relate to how the echoes are received and processed rather than the pulse transmission. Signal processing (also known as speckle reduction) techniques are designed to interrogate lines of echoes obtained on each frame and improve the signal-to-noise ratio by pattern recognition [5]. For example, an anechoic area containing background noise will be recognised as a cystic structure and the abrupt change in echoes around the margins of this target then recognised as the cyst/vessel wall (e.g., Precision Imaging^TM, Toshiba). This process of removing unwanted noise leads to images that look sharper because of the improved signal-to-noise ratio and emphasise cystic vs. solid consistency. The dynamic range and grey-scale map are inherently linked but different aspects of high-resolution ultrasound. The dynamic range controls the range of grey (or the preferred colour tint) that is available to display echo intensity, whilst the grey-scale map allocates which parts of this spectrum are used to represent the echo signals received [6]. Tissue harmonic imaging (THI)—an established technique for improving image contrast at other body sites—has recently become more relevant to high-frequency ultrasound. As the basis of this technique is to receive a signal from the second harmonic frequency of the probe’s transmit pulse (i.e., twice the probe transmit frequency), its use in superficial structures has been limited [7]. However, the advent of wider bandwidths and more sophisticated receive technology means that THI is increasingly available to use with high-frequency probes (e.g., Hitachi High definition dynamic THI) resulting in further removal of clutter/noise and improved tissue contrast. Basic ultrasound settings (focal zones, time gain control, etc.) also affect image quality but are not covered in this article.

An abnormal thyroid nodule margin is a recognised feature of thyroid malignancy. Assessment of the nodule edge has previously been binary; the margin is either well or poorly defined. A poorly defined/irregular nodule margin conferred 87 % specificity for malignancy in a study population of 1,108 thyroid nodules but with a very low sensitivity of 39 % [8]. Whilst operator experience is an important factor in edge delineation, the machine settings heavily influence this assessment and create a challenge for standardised reporting between observers using different equipment. The probe frequency, bandwidth and use of compound imaging are the key technical aspects to consider here (Fig. 2), delivering improved signal-to-noise ratios for the operator to delineate the nodule from the surrounding thyroid parenchyma and adjacent structures. More recently, the concept of a ‘spiculated’ margin (Fig. 3) has been introduced to describe the finding of a distinctly infiltrative nodule edge (as separate from the broader ‘poorly defined’ category) with 92 % specificity and 81 % positive predictive value in a comprehensive review of B mode ultrasound findings [9]. The ability to detect margin spiculation is a consequence of higher spatial and temporal resolution and it therefore follows that the reliable detection of this useful characteristic requires access to the advances in the above-described ultrasound technology. Understanding the close relationship of technology to sonographic interpretation is particularly important because—of all the individual thyroid nodule features—margin delineation has the poorest inter-observer agreement [10]. The multiple technical factors are likely to contribute to this variability and are particularly relevant when patients undergo thyroid ultrasound at different centres with varying ultrasound probe and software settings.

• Smaller nodule detection:

The rising incidence of thyroid cancer across many countries is almost entirely attributable to small (<2 cm) papillary carcinoma [11]. Multiple factors have contributed to the phenomenon of the clinically occult papillary thyroid cancer; incidental detection on other imaging modalities, patient awareness and access to health care amongst others. The ability to visualise and identify suspicious features within these small nodules (e.g., margin irregularity, microcalcification) is, however, a consequence of the advances in high-resolution ultrasound. Sub-centimetre nodules can be identified, characterised and targeted for FNA if appropriate (Fig. 4). This has created difficulty for guidelines that have been primarily based on nodule size (i.e., the Society of Radiologists in Ultrasound [12]), particularly given the increasing understanding that size alone is not an accurate predictor of malignancy [13]. Moreover, improved resolution enables better needle tip visualisation during US-guided FNA/biopsy and obtaining an adequate cytological specimen from very small (i.e., 5 mm) nodules is feasible, although lower adequacy rates are reported from FNA of these smallest lesions [13]. The clinical benefit of confirming malignancy in this setting is controversial—and beyond the scope of this article—as the prognosis for small papillary thyroid cancers is generally excellent, prompting the proposal of a separate nomenclature, ‘papillary lesion of indolent course’ [11]. The reality for the thyroid US operator at present is that he or she will regularly be confronted with small but suspicious incidental nodules that would previously have remained undetected.

• Nodule consistency

Most thyroid cancers are hypoechoic compared to the background parenchyma. Unfortunately, this finding is non-specific as benign nodules can also be hyporeflective and the comparison with ‘normal’ thyroid may be hampered by background nodularity or thyroiditis. By comparison, marked hypoechogenicity—defined as ‘less reflective than adjacent strap muscle’—is a specific malignant finding with an odds ratio of 8.46 in a retrospective study of 849 thyroid nodules [14]. Although this feature is only seen in the minority of cancers, its recognition makes a benign nodule unlikely and will typically initiate further investigation with FNAC (Fig. 5). Dynamic range and grey-scale map settings predominantly control image contrast whilst the use of harmonic imaging will further emphasise differences in reflectivity. A narrow dynamic range will increase the contrast between adjacent structures by utilising fewer shades of grey to display the echo signal intensity. Whilst these settings are typically configured within manufacturer presets, they can be altered easily on most ultrasound platforms and should ideally be customised to the user’s preferences. The clinical value of marked hypoechogenicity as a specific sign of thyroid cancer emphasises the importance of being familiar with one’s own machine settings.

• Echogenic foci—colloid vs. calcium

In addition to the choice of the dynamic range and grey-scale map, signal processing techniques help to identify small hyperechoic foci. Bright spots that show a ‘ring down’ or ‘comet tail’ artefact are a very helpful finding; they represent foci of inspissated colloid and are a feature of benign nodules. Conversely, the presence of sub-millimetre, highly reflective foci with or without acoustic shadowing represents microcalcification—a specific feature of differentiated thyroid malignancy [15]. The depiction of the comet tail artefact is therefore crucially important to confirm the benign nature of an echogenic focus. As would be expected, signal processing emphasises the bright focus and can help to make the comet tail more conspicuous to the operator (Fig. 6).

Traditionally, the detection of microcalcification (usually due to small psammoma bodies within papillary thyroid cancer) was made easier by the associated acoustic shadow artefact created by reflection of the perpendicular ultrasound beam [16]. In high-resolution ultrasound, the use of spatial compounding means that the signal can be received from tissue deep to calcific foci (Fig. 7) and this artefact is often diminished or lost in the thyroid. This unwanted consequence of modern technology can be overridden by deactivating spatial compounding and allowing acoustic shadows to emerge—if the operator understands the significance of the artefact and its causation.

An alternative method of highlighting microcalcification within thyroid nodules is to alter the grey-scale map, dynamic range and speckle reduction combination (Fig. 8). Boosting the brightness of the signal received from calcific spots relative to normal parenchyma increases the detection and has been investigated in breast ultrasound (where the identification of microcalcification also carries significant clinical value) [17]. A novel modified grey-scale map/image tint combination to detect microcalcification (MicroPure Imaging, Toshiba) has been recently reported in thyroid nodules [18].

The technological progress of B mode US enables the thyroid gland to be assessed in more anatomic detail than ever before, characterising lesions by a matrix of structural features. Building on these advances, the value of three-dimensional US in thyroid nodule assessment has recently been considered [19–21]. Initial studies have focussed on two aspects: the ability to more accurately assess the key B mode features (e.g., nodule margin) and the potential to uncover independent predictors of malignancy from 3D data sets. Whilst 3D US is not widely used in the neck at present, recent developments in this area suggest the potential to address persisting challenges of thyroid imaging.

In a prospective study of 91 thyroid nodules, the interobserver agreement for specific characteristics (e.g., shape, margin) was better for 3D than 2D US. Although intra-observer agreement remained suboptimal, there was significantly stronger agreement between users for suspicion of malignancy after off-line review of the static 3D images [21]. National guidelines are increasingly based on standardised classification of malignant risk [1, 22] (as opposed to subjective assessment by experienced individuals); therefore reducing inter-observer variation is particularly relevant [1, 21, 22].

In a retrospective study of 71 thyroid nodules, 3D ultrasound data were analysed using multiplanar reformatted (MPR) images and a thin-slice volume-rendered technique [19]. The MPR data found that poorly defined 3D nodule margins were associated with malignancy but also described two novel, independent 3D predictors of malignancy: a lobulated nodule shape in the C-plane (coronal) and altered central 3D vascularity. The smooth surface, thin-slice volume-rendering algorithm created images with higher contrast and reduced noise: a new layer of post processing to compare with existing 2D techniques. The shift in emphasis from real-time image interpretation to volume acquisition with subsequent processing and multiplanar viewing also offers potentially more detailed data analysis. A study of 20 thyroid nodules assessed 3D high-resolution ultrasound data with a computer-aided diagnostics (CAD) software programme [20]. The authors found that CAD-detected specific 3D textural features could be identified and combined as an accurate discriminator of benign and malignant nodules, creating automated malignant risk stratification (‘Thyroid Malignancy Index’). Such an approach would represent a paradigm shift in practice for most thyroid imagers but the potential for accurate objective data to support an operator’s skill and experience is attractive.

Regarding contrast-enhanced ultrasound (CEUS), a small number of studies have documented the feasibility and diagnostic accuracy of assessing thyroid nodule enhancement following intravenous microbubble contrast, using low mechanical index (MI) settings [23, 24]. Recognisable quantitative patterns of perfusion have been identified to construct CEUS malignant risk stratification with 76.9 % sensitivity and 84.8 % specificity in a study of 42 patients [24]. These findings suggest a potential role for CEUS as a diagnostic adjunct to B mode US but the technique is more invasive and time-consuming than standard ultrasound and, at present, the specific clinical utility is unclear.

Elasticity refers to the tendency of a tissue to undergo a reversible deformation and is quantified by Young’s modulus (E) in kiloPascals (kPa), which equals the ratio of an applied stress to the induced strain (displacement/original length) [25, 26]. USE technologies either estimate elasticity or a surrogate parameter such as strain from US signals acquired during a mild tissue deformation, which can be displayed two-dimensionally as elastograms. A range of proprietary USE technologies is available commercially, which can be divided into strain elastography (strain USE) and shear wave elastography (SWE) based on their underlying physics.

In strain USE, also known as quasi-static elastography, compression elastography and real-time elastography (RTE), a mechanical stress is used to deform the tissue of interest and returning US signals are software processed to generate dynamic elastograms of relative tissue strain. In the thyroid, either the operator applies freehand compression to gently compress and decompress a nodule axially under the transducer, or alternatively nodules in the lateral lobe may be compressed solely by transmitted pulsations from the adjacent common carotid artery (Fig. 9). Elastograms are displayed as an opacity layer over and beside corresponding grey-scale US images (split screen mode) using a chromatic scale that varies between manufacturers. Operators performing freehand compression must use a precise manual technique to generate high-quality elastograms, which can be optimised by referring to compression quality scales that are automatically computed and displayed in real time (Fig. 10). Strain elastograms are either interpreted qualitatively, based on visual grading of the proportion and distribution of different colours within the nodule (Fig. 11), or semi-quantitatively using strain ratios (computed from regions of interest (ROIs) placed in the nodule and adjacent thyroid parenchyma) or other indices such as strain heterogeneity. Presently, there is no agreement on an optimal qualitative scoring system although many investigators have used 4- or 5-point scales whereby higher elastography scores (ES) signify higher nodule stiffness [27, 28]. Strain USE is unsuitable for nodules without sufficient reference tissue in the elastogram, which basically excludes single or conglomerate nodules exceeding ~3 cm diameter.

SWE technologies applicable to thyroid imaging use highly focused ultrasound impulses to induce minuscule localised lateral displacements at different locations within the tissue called shear waves, followed by ultrasound detection impulses to track shear wave propagations [25]. SWE is also known as acoustic radiation force impulse imaging (ARFI). Shear waves travel faster in stiffer tissues, whereby shear wave velocity (SWV) is directly proportional to the square root of Young’s modulus. Unlike strain USE, SWE output is quantitative; expressed as SWV (m/s) or estimated tissue stiffness (kPa). Due to differences in proprietary SWE technologies between manufacturers, SWE outputs are varied. These include static colour-coded elastograms termed 2D SWE or a single numerical estimate for an ROI of fixed dimensions (~5 × 6 mm) termed point SWE (pSWE). At least one commercial SWE system produces elastograms in real time, from which SWV or Young’s modulus estimates can be computed for ROIs selected by the operator (Fig. 12).

• Accuracy of USE for malignancy

Presently about 80 studies of strain USE and 20 studies of SWE have been published for thyroid malignancy since 2005 [29–35]. Most are single-institutional series with heterogeneous designs, comprising around 100 nodules selected for FNAC or surgery because of suspicious or equivocal sonographic features, suspicious or malignant cytology or nodules within a compressive goitre. Most studies have also variably excluded nodules containing coarse macro-calcifications, substantial cystic areas or those within a diffuse thyroiditis as early evidence suggested that USE is suboptimal in these cases. Due to this selection bias, a much higher proportion of nodules in these studies is malignant (~25 %) than in an unselected population (~5 %) and most are papillary carcinomas (~90 %); thus published discriminatory performance data actually refer to this histological type [29–35].

Promisingly, most studies indicate that papillary cancers have significantly higher stiffness indices on USE compared to benign nodules although there is some overlap (Figs. 11, 12, 13 and 14). Furthermore, most evidence suggests that USE has a superior overall accuracy compared to single or multiple conventional US criteria. Multiple meta-analyses confirm these findings [29–35] including the largest to date, which reports a pooled mean sensitivity, specificity, PPV, NPV and accuracy of 87, 80, 47, 97 and 81 % for strain USE using freehand compression (N = 4,926); 87, 80, 52, 100 and 82 % for strain USE using carotid artery pulsation (N = 241) and 86, 89, 60, 97 and 89 % for SWE [34]. From the pooled data, SWE has higher accuracy compared to strain USE, while performing strain USE using carotid arterial pulsations is slightly superior diagnostically compared to freehand compression. The precise cause for the high stiffness of papillary cancers needs to be clarified but probably reflects intense tumour fibrosis, with or without calcifications, which is characteristic of this histology although not always present (~80 %) [36]. Indeed, anecdotal examples of high USE indices are reported for benign nodules containing dense fibrotic foci and calcifications histologically as well as low USE indices in some papillary and follicular cancers lacking substantial fibrosis [37].

USE is a dynamic technique with fluctuating elastographic appearances due to its intrinsic sensitivity to miniscule displacements that are continually occurring under physiological conditions. Various components of USE acquisition and interpretation are operator dependent including freehand compression for strain USE, selection of imaging planes through nodules, selection of representative elastograms, qualitative scoring of elastograms and placement of ROIs for (semi-)quantitative measurements [78]. Strain USE using freehand compression is highly dependent on an optimal axial compression technique, which is hampered by competing non-axial motions (e.g., arterial pulsations and respiration) and the different mobilities of tissues bordering the thyroid (e.g., immobile trachea and mobile vessels). Carotid artery strain USE and SWE circumvent this source of operator dependence. Nevertheless, irrespective of the USE technology used, the operator influences the amount of resting pressure applied via a motionless transducer, termed precompression, which in turn can alter tissue stiffness [34]. This phenomenon reflects the fact the Young’s modulus of biological tissues increases as strain is increased, termed strain hardening, although this effect is minimal for very small strains (<10 %). To mitigate this problem, USE should be performed with minimal pressure applied, which may be facilitated by maintaining an ample US gel layer between the transducer and skin.

Various elastographic pitfalls and artefacts can occur in thyroid USE, which need to be recognised and avoided. High stiffness areas due to the stress concentration frequently develop in the near field beneath the transducer including the superficial aspect of some thyroid nodules, as well as at boundaries of tissues with contrasting stiffness including the margins of some nodules. Isthmic nodules are especially prone to stress concentration effects because they tend to produce a focal skin bulge with uneven contact under the flat transducer and be compressed directly against the unyielding trachea. Other artefacts can also develop because of a suboptimal compression technique, competing non-axial motions, presence of cystic fluid and limitations of specific USE technologies (Fig. 15) [25, 79].

• Reproducibility of USE

Given the aforementioned sources of operator influence, practical challenges and artefacts, establishing a high reproducibility of thyroid USE is an important objective. An early report of strain USE had disappointing reproducibility results [80]; however most recent studies using newer USE technologies report substantial to excellent intra- and inter-observer agreements [49, 55, 57, 79, 81–85]. Evidently, the inclusion of compression quality assurance scales on strain USE systems has had a positive influence on this aspect. Nevertheless, the fact remains that (semi-)quantitative USE indices have differed substantially between studies, including those evaluating identical SWE technologies [35]. These discrepancies may reflect sampling variations although their magnitude and pattern raises questions about additional mechanisms, which in turn may influence reproducibility. One postulation is that operators may be performing USE using different degrees of precompression (Fig. 16) and may even be unconsciously influenced by the nodule's grey-scale appearances on the split-screen display (cognitive bias) [86, 87]. At present, precompression cannot be measured or standardised reliably because pressure sensors are not incorporated within transducers.

• Current and future directions of USE

According to recently published guidelines from the European Federation of Societies of Ultrasound in Medicine and Biology (EFSUMB): “elastography is an additional tool for thyroid lesion differentiation” and “based on expert opinion, elastography may be used to guide follow-up of lesions negative for malignancy at FNAC” [88]. This cautious endorsement reflects the current controversies in thyroid USE evidence. Clearly further research is required including large multi-centre prospective trials of nodules with less pre-selection to evaluate different nodule characteristics, uncommon pathologies and reproducibility and ultimately determine how USE can be integrated with conventional US into emerging malignant risk classification systems (e.g., TIRADS). In this respect, clarifying precisely which types of nodule are unsuitable for thyroid USE will be critical to optimising its accuracy.

Although controversial, USE shows promise as an adjunct to conventional US for reliable identification of benignity in a select group of nodules on the initial US examination as well as those with non-diagnostic or indeterminate cytological results following FNAC. Even if these indications are validated in future studies, several issues will still need to be addressed before USE can be accepted widely in routine practice. USE technologies should be robust and simple to apply by the range of health care professionals who perform thyroid US routinely. The specific strengths and limitations of each USE technology including artefacts need to be catalogued systematically. While it is acknowledged that USE technologies are proprietary and their outputs may not be interchangeable, there should be greater standardisation of thyroid USE, for example by unifying nodule-grading systems and elastogram chromatic scales. Fortunately, USE technologies are continually improving including with respect to measurement precision and quality assurance, which should augment their diagnostic accuracies and reliabilities. Ideally, with advances in CAD software, a nodule’s grey scale and elastographic features could be analysed automatically and converted into a single estimate of its malignant risk. There are still some hurdles to overcome although, at the current pace of promising research and technological advances, USE may well become a valuable complementary technique for evaluating thyroid nodules in the near future.