The two main alternatives to CT to consider for imaging bone of the head and neck region are ultrasound and MRI. Other techniques such as optical coherence tomography which can evaluate the microstructure of bone, are beyond the scope of this review.
Ultrasound provides excellent soft tissue imaging of the head and neck. Whilst it is not typically the modality of choice to evaluate bony structures, the cortical bone surface is easily delineated, making superficial pathologies such as osteomas easy to diagnose. Cranial ultrasound is frequently employed in the paediatric setting, permitting evaluation of bone, cranial sutures and intracranial contents.
In infants, cranial ultrasound provides both excellent visualisation of bony pathologies with intracranial assessment being afforded by patent fontanelles and often minimal hair. In the investigation of abnormal head shape in infants, ultrasound has been shown to offer comparable accuracy to CT in the diagnosis of craniosynostosis. Rozovsky et al.  evaluated the sensitivity and specificity of ultrasound compared to radiography in infants investigated for main sutural craniosynostosis. Ultrasound performed comparably to radiography with a sensitivity of 100% and specificity of 98%. Discrepancy occurred in two infants for the metopic suture in whom physiological closure of the suture had taken place, being mistaken for synostosis on ultrasound. Interpreting the ultrasound findings in unison with the clinical appearances would have avoided this diagnostic error—the absence of trigonocephaly providing significant reassurance that the suture was not prematurely fused. Similar results were reported by Alizadeh et al.  in their study of 44 infants under the age of 1 year with clinical suspicion of craniosynostosis. Sensitivity and specificity of ultrasound was 96.9% and 100% respectively.
The main difficulty associated with using ultrasound for diagnosis of craniosynostosis is the inability to produce 3D reconstructed imaging of the whole cranium, which many surgeons require for surgical planning.
In the investigation of traumatic skull fractures, Rabiner et al.  demonstrated a sensitivity of 88% and specificity of 97% in a cohort of 69 patients with a mean age of 6.4 years. Similar results were obtained by Parri et al.  in a cohort of 115 patients under the age of 2 years.
The evidence for point of care ultrasound for the diagnosis of skull fractures in children has been recently reviewed in a systematic review and meta-analysis by Alexandridis et al. [12* *], who appraised 7 relevant studies. Sensitivity ranged from 67–100% and specificity from 85%-100%. Similar findings were reported in a further systematic review by Gordon et al. .
Good intracranial views can often be achieved with ultrasound, particularly during the neonatal period, which permits completion of the examination. Masaeli et al.  compared the ultrasound and CT findings of 538 patients under the age of 18 years. The point of care ultrasound was performed by attending emergency physicians and third year residents with a sensitivity of 85.7% and specificity of 97.9% for the detection of intracranial haemorrhage in children under 2 years of age, with good concordance with CT findings. There was however 1 false negative and 3 false positives in this age group.
In the trauma setting, the ability to perform bedside ultrasound in the emergency department, is highly beneficial, offering a rapid mode of assessment, without the need for ionising radiation. Consideration does however need to be given to the mechanism of injury and the potential of abusive head trauma (non-accidental injury) which will necessitate full investigation with cross-sectional imaging to fully document all injuries.
As with all ultrasound examinations, there is reliance upon operator expertise, with the potential for misinterpretation, patient non-compliance in the paediatric setting and the inability to provide a clear surgical roadmap for the surgeon. Increased reliance is placed on the ultrasound report due to unfamiliarity of ultrasound images by the clinical team, and is notoriously disliked for this reason. Frequently, point of care ultrasounds are performed by the resident clinical team, with wide variability in ultrasound experience and expertise. In the meta-analysis by Alexandridis et al. [12* *], ultrasounds were typically performed by clinicians after a training period of 1 h. This will contribute to the reported specificity of the various studies, with operators with the most experience and training invariably attaining the most accurate diagnoses.
The main contender for bone imaging therefore lies with MRI, which will be the focus for the remainder of this review. In recent years techniques have developed to offer volumetric acquisition with isometric voxels. However, MR depiction of solid bone structures is challenging due to low proton density (~ 20% of water) and heterogeneous structure, which leads to short signal lifetimes . Techniques to enhance bony detail and create 3D rendered images is an area of ongoing research, in which considerable progress has been achieved over the past 14 years.
MRI Bone Techniques
Black Bone (GRE-BB)
Tissues such as cortical bone have tightly bound protons, resulting in local magnetic field heterogeneity and short spin–spin (T2) relaxation times (< 1 ms), leading to weak MR signal which decays very rapidly. The relatively long echo times (TE) used in conventional imaging will produce little or no signal and appear dark (or “black”). Whilst tissues such as cortical bone, tendons, ligaments, and menisci contain a majority of short T2 components, other tissues also contain short T2 relaxation components, but as a minority species .
The first reported MRI technique in the literature for imaging bone of the craniofacial region was referred to as “Black Bone”, acknowledging this short T2 effect. By manipulating imaging parameters to make soft tissue intensities as uniform as possible (i.e., approaching proton-density weighting), the bone soft-tissue interface is enhanced  (Fig. 1). The initial descriptions of Black Bone utilised a gradient echo sequence (GRE-BB), with short TE, short TR and a flip angle of 5 degrees. With a 3D volume acquisition, a voxel size of approximately 1.2 mm is achieved with a scan time in the region of 5 min for an adult head. Biometric accuracy has been confirmed through phantom work . Whilst the initial imaging was performed using GE scanners, translation to other vendors produced similar results with Philips and Siemens MR systems [18**, 19]. The technique relies on a gradient echo (GRE) sequence such as 3D fast GRE (GE Medical Systems Ltd, Chalfont St Giles, UK), volumetric interpolated breath-hold examination (Siemens Healthcare Ltd, Camberley, UK) or an equivalent T1 weighted GRE 3D volume sequence (Philips Healthcare, Guildford, UK). The main limitation of the technique is in areas where air and bone come into contact, since the two species essentially yield no signal and therefore cannot be distinguished. The main advantage, however, is that this simple technique can be performed on all magnets without requiring any additional sequences or hardware.
Over recent years, the term Black Bone has been used interchangeably for a variety of MRI techniques whereby the goal remains the same, yet the fundamental physics can be considerably different.
A family of clinically compatible sequences is capable of achieving TE values less than 1 ms, including ultrashort echo time (UTE) and zero echo time (ZTE) sequences . In the case of UTE, short excitation pulses followed by radial readout are used to minimise the time between excitation and the acquisition of the k-space centre. This results in echo times of a few tens of microseconds, capable of recording the rapidly fading signal from cortical bone. One important advantage of detecting bone signal, however weak, comes for automated segmentation. Indeed, by combining two or more acquisitions with different echo times, it is possible to identify and separate long T2 species from cartilage and bone.
A further reduction of the dead time between excitation and readout can be achieved by ramping up the encoding gradients before, rather than after, the excitation pulse. With this approach, used in sequences such as Pointwise Encoding Time Reduction with Radial Acquisition (PETRA) and Zero Echo Time (ZTE), echo times are mainly limited by the hardware’s ability to switch between transmit and receive modes. The high sampling efficiency of close to 100% (i.e., most of the TR is used to acquire data) provides efficient scanning and offers a flexible trade-off between bandwidth, number of averages and signal-to-noise ratio . Indeed, the signal-to-noise improvement brought by the shorter echo times enables the segmentation of bone structures from a single acquisition, with the subsequent time gain with respect to dual-echo UTE approaches.
As a 3D acquisition, these sequences can obtain isotropic resolution with 1 mm slice thickness, with an acquisition time of approximately 4 min for the head (Fig. 2). A significant impetus for the development of ZTE sequences was for attenuation correction in PET/MRI, but ZTE also shows considerable potential in imaging a range of bony pathology.
Conventional MRI sequences result in acoustic noise in the region of 90–130 dBA due to Lorentz forces caused by rapidly changing currents in the magnetic field gradient coils primarily used for spatial localisation . This is a significant contributing factor in failed MRI examinations, particularly in the paediatric population (where imaging during natural sleep is often attempted), with resultant motion artefacts or premature termination of the study. However, with ZTE, gradient switching can be reduced to a minimum which enables quiet operation, being highly beneficial for paediatric imaging . The main limitation of ZTE techniques is the hardware requirements, such as high-performance coils permitting rapid switching between transmit and receive. As such, UTE/ZTE cannot be adopted on all scanners.
CT-like and Synthetic or Pseudo-CT Imaging
Various techniques have been utilised to generate CT-like images from MRI datasets (Fig. 3). Two broad sub-types exist – those that mimic the visual appearance of CT images and quantitative synthetic CT images that are based on the accurate approximation of measurable Hounsfield units . Utilising a deep learning-based multiparametric MRI technique, Jans et al.  were able to create synthetic CT images that qualitatively and quantitatively resemble conventional bone window CT images.
3D Reconstruction Techniques
The first reports of 3D reconstruction of the craniofacial skeleton from GRE-BB were presented at the CARS conference in 2011 , with subsequent publications in 2014 focused on children with craniosynostosis [18**]. The GRE-BB sequence was successfully utilised to 3D render the craniofacial skeleton using Fovia High-Definition volume rendering software (Fovia Inc, Palo Alto, CA, USA). This necessitated the selection of an appropriate transfer function, whereby a colour and opacity were applied to every voxel, followed by segmentation to remove the soft tissues. Even with some semi-automated processing, the average time for image processing was in the region of 13 min per dataset in experienced hands. Further results utilising mainstream software such as Mimics (Materialise, Leuven, Belgium), and Osirix (Pixmeo, Switzerland) were also described, with varying results . Fully automated segmentation algorithms have now been achieved for adult GRE-BB datasets, with ongoing work to develop similar results in young infants  (Fig. 4). For automated segmentation, a series of steps are required to produce pseudo-CT images with the bone appearing hyperintense in comparison to surrounding tissues. It is not possible to simply invert the image to achieve this, since air and bone have similar pixel values, and thus the head would be obscured by a box of hyperintense air included in the field of view.
In 2016, Wiesinger et al.  demonstrated the potential of ZTE for 3D reconstruction of the skull. The high resolution ZTE sequence took 6 min 12 s to acquire images of the head in their adult volunteers. Following bias correction and signal normalisation, the images were segmented into three classes: air, soft tissue and bone, and the pseudo-CT imaging successfully 3D rendered. Lu et al.  imaged 14 patients (2 months to 17 years of age) with ZTE for a range of clinical indications, the majority of which were skull deformity either congenital or traumatic. Most patients were imaged with both CT and ZTE, and the authors reported similar success with 3D rendering.
Other MRI Techniques
Maeda et al.  utilised a 3D T1-weighted sequence to surface render the mandible, and compared the results to CT. They reported a 3D absolute difference between CT and MRI of 1.6 mm. The use of a T1-weighted sequence will have contributed to this discrepancy, in view of the greater challenges and reliance upon manual segmentation for a non-bone specific MRI technique.
Combining MRI Techniques
By utilising the main advantage of ZTE, namely that it is the optimal method to discern air from bone and using it with another sequence such as GRE-BB, to provide high resolution imaging without the inefficiency of radial sampling, 3D segmentation algorithms are further enhanced [29*]. This helps to minimise non-bone containing structures, such as ligamentous attachments as well as air. These technical developments, utilising ZTE as the driver, also permits the use of alternative MRI sequences, such as FIESTA-C (fast imaging employing steady-state acquisition; GE Healthcare), which offers high resolution imaging but with slightly shorter acquisition times (Fig. 5). FIESTA-C is used more routinely in clinical practice, and thus widens the scope of clinical utility, since only the relatively rapid acquisition of ZTE imaging would be required to permit 3D bone reconstruction; low-resolution ZTE imaging of the head being acquired in little over one minute.
Utilising FIESTA-C imaging to produce 3D images of the skull, and time of flight imaging of vessels Hayashi et al.  were able to successfully create 3D visualisation of the frontal and parietal branches of the superficial temporal artery, in patients undergoing investigation for ICA-ECA bypass. This provided superior evaluation of the vessels than CT due to beam hardening of the skull.
Bone-specific MRI techniques have largely been focused on paediatric pathology in view of the greatest potential in terms of radiation protection.
Cranial Sutures and Craniosynostosis
The earliest work in this area was in children with craniosynostosis. GRE-BB successfully depicted the normal cranial suture as an area of hyperintense signal, distinct from those which were prematurely fused and thus not visualised [18**] (Fig. 6). GRE-BB offers enhanced detection of the unfused cranial sutures to ZTE. Cranial suture detection was further enhanced by the 3D reconstruction work previously discussed [18**]. Fully automated segmentation algorithms have recently been applied to these datasets with variable results (Fig. 7). Ongoing research in this area is likely to yield a fully automated segmentation for young infants. Patel et al.  utilised a Siemens GE-VIBE (fast low-angle shot golden-angle 3D stack-of-stars radial VIBE) base sequence on a 3 T magnet, with similar TR/TE and a flip angle of 3 degrees. Non-isotropic voxels were obtained (0.6 × 0.6 × 0.8 mm), with a similar acquisition time of 5 min 4 s. 14 participants were recruited, but imaging could not be acquired in 2 participants due to non-compliance, and a further dataset was severely degraded by motion artefact. The age range for the available MRI datasets was 3 weeks to 9 years (median 1.8 years). Two examples of the 3D MRI rendering were provided, requiring an MRI processing time between 30 min and 2 h per patient. The long image processing time make this technique prohibitive. Fully automated segmentation techniques are a necessity for these techniques to finally be routinely adopted into clinical practice.
Dremmen et al.  studied the imaging of 28 children with acute head trauma, where CT and MRI (with GRE-BB) imaging had been acquired within 7 days of each other. GRE-BB imaging showed a sensitivity of 66.7% and a specificity of 87.5% in the detection of skull fractures, utilising axial acquisition with reconstruction in the coronal and sagittal planes. Two false negative cases were misinterpreted as cranial sutures, and similarly two false positives also related to misinterpretation of sutures as fractures. A further false negative was due to misinterpretation of a fracture as a venous transosseous vascular channel. These cases highlight that with increasing use of the GRE-BB technique, familiarity with normal will be increased, and thus specificity improved. The one case where this would not improve was in a case where the fracture was in the mastoid region. Air and bone are almost impossible to discern on GRE-BB—one of the main limitations of the technique, which may be overcome by the addition of ZTE.
It is well recognised that the addition of 3D reconstructed images from CT increases the sensitivity and specificity for detecting skull fractures and is essential in the evaluation of paediatric head CTs for distinguishing subtle fractures from sutural variants, especially in the setting of trauma . It is therefore unsurprising that the addition of 3D reconstructed MR images of the skull in children with abusive head trauma offered improved sensitivity and specificity compared to Dremmen et al.  study. PETRA imaging (ZTE) was successfully utilised in children between the ages of 1–30 months with semi-automated segmentation (Fig. 8). This yielded a sensitivity of 83% and specificity of 100% for skull fractures when compared to conventional CT imaging . Similarly, in an adult population, Cho et al.  found that ZTE was equivalent to CT in the diagnosis of skull fracture in 13 patients, and that there was no significant difference in the measured skull thickness on both modalities.
Temporal Bone Imaging
MRI is typically the primary imaging modality in pre-operative cochlear implant assessments, providing the ability to assess cochlear nerve calibre and the patency of the cochlea. However, visualisation of bony anatomy, such as the mastoid facial nerve canal, are better served with CT. Connor et al.  were able to successfully delineate the mastoid facial nerve canal in all 25 adult patients, and the posterior canaliculus of the chorda tympani in 72% of cases. The authors concluded that performance was comparable to CT, and thus the addition of a GRE-BB technique would be valuable in identifying these landmarks and thus defining the facial recess.
Temporomandibular Joint Imaging
Lee et al.  evaluated the clinical usability of ZTE in assessing the bone changes of the temporomandibular joint, in comparison to CBCT. They found high reproducibility for ZTE, comparable to CBCT. The addition of a ZTE sequence to routine MRI examination would obviate additional CT imaging in this cohort.
Cephalometric analysis was introduced in the early 1930’s and is a key element of orthodontic and orthognathic surgery. Traditionally measurements have been performed on two-dimensional radiographs. The limitation of this 2D representation of 3D anatomy led to increased use of CBCT or conventional CT. Justifying this additional radiation exposure is difficult.
The potential application of GRE-BB for cephalometric analysis was unveiled at the European Society of Head & Neck Radiology in 2010, and subsequently reported in the literature in 2013 . This opened the door to potential radiation-free methods of 3D cephalometry. The main limitation of course remains that such analyses are frequently required in patients with fixed orthodontic appliances, for which a gradient echo sequence will result in significant metallic artefact. Subsequent research in this area has included the use of a T1-weighted 3D MSVAT-SPACE (multiple slab acquisition with view angle tilting gradient based on a sampling perfection with application optimized contrasts using different flip angle evolution) prototype sequence to perform 3D cephalometry .
Head & Neck Malignancy
MRI provides superior soft tissue imaging and delineation of tumours compared to CT. Staging of certain head and neck cancer subsites includes bone invasion as a reason for upstaging using the TNM system. Whilst marrow signal change is more easily detected on MRI, cortical erosion on standard sequences can be challenging. The addition of GRE-BB or ZTE permits improved visualisation of the cortical bone often obviating the need for CT imaging (Fig. 9 and 10). Since CT is frequently affected by dental amalgam artefact, this technique can be particularly valuable . Suzuki et al.  evaluated 49 cases of suspected mandibular invasion comparing a conventional 2D FSE and high-resolution 3D VIBE with CT. All three techniques correctly diagnosed the 32 tumours with histopathological evidence of mandibular invasion (sensitivity 100%). The specificities of 2D FSE, 3D VIBE and CT were 56, 78 and 89%. It should be noted that the flip angle for VIBE was 20 degrees, and the TR/TE 13.9/3.9, and thus not sufficient to minimise the soft tissue contrast to produce Black Bone images. Further research in this area is required to determine if GRE-BB or ZTE can achieve a higher specificity for cortical erosion.
Tumour volume has previously been demonstrated to be a better predictor of survival in oral cavity cancers than linear measurements, with Black Bone being successfully utilised to perform volume measurements in a range of cases .
PET Attenuation Correction
Typical MRI-based attenuation correction methods for PET imaging use a fat/water map derived from a two-echo Dixon sequence, where bone is neglected. Leynes et al.  in 2017 proposed the use of patient-specific multi-parametric MRI consisting of Dixon MRI and proton-density weighted ZTE to directly synthesise pseudo-CT images with the use of a deep learning model. Wiesinger et al.  utilised proton density-weighted ZTE images to produce pseudo-CT which provided comparable attenuation information to CT. Delso et al. [43**] demonstrated excellent correlation between ZTE and CT bone segmentations, concluding that single-echo proton density-weighted ZTE is an efficient means of obtaining anatomically accurate maps of bone tissue for the purpose of PET attenuation correction.
Treatment planning systems rely on electron density maps generated from CT images for calculating dose. Lei et al.  developed a learning-based method to generate patient specific synthetic CT from a routine anatomical MRI for use in radiotherapy planning. The algorithm was evaluated in 14 patient datasets consisting of MRI and CT, demonstrating the feasibility of MR based dose calculation.
Head and neck imaging is frequently acquired not for diagnostic purposes but for surgical planning of complex cases. Several authors have evaluated various MRI Black Bone techniques for this purpose. This includes CAD/CAM planning in children with non-syndromic metopic synostosis with successful production of cutting guides . The authors also highlighted the added benefits of MRI acquisition in this patient cohort, namely superior illustration of the neurovascular and dural venous system, and brain anomalies .
Hoving et al.  developed a 3D virtual planning workflow based on GRE-BB for the lower jaw. The most challenging area to segment, as expected, was the coronoid process in view of the dense muscular attachment, but an area not of interest in most clinical settings.
3D printing is an integral component of surgical planning for a wide range of craniomaxillofacial surgical procedures. Once a 3D imaging dataset has been fully segmented, the process of creating a suitable file for 3D printing is relatively straight forward. The first 3D printed GRE-BB model was of an adult mandible created on a Z650 binder and powder 3D printer . Utilising a cube phantom, it was confirmed that 3D printing from GRE-BB datasets demonstrated a similar level of accuracy to CT imaging datasets.
Recently, Low et al.  reported their experience of using GRE-BB for a range of head and neck pathologies in children. This included pathologies already discussed, and a range of focal osseous/fibro-osseous lesions such as fibrous dysplasia and Langerhans cell histiocytosis (LCH). Other reported applications include the assessment of facial asymmetry, benign tumours of the jaws, and facial fractures . It can also assist in making salivary calculi more conspicuous (Fig. 10).
Limitations of MRI
The main limitation of MRI is the potential for motion artefact, which can be problematic in the paediatric population. However, in a recent study Jaimes et al. reported an overall success rate of 82% for non-sedated MRI examinations in children aged between 1–7 years . As expected, young age impacted upon success. Feed and wrap methods are typically only successful in infants less than 3 months of age. However, Dean et al.  reported a success rate of 97% in children under 4 years of age undergoing MRI examination during natural non-sedated sleep. The use of silent ZTE techniques are particularly valuable when imaging children, and whilst GRE-BB is not a silent sequence, the noise that is generated is frequently well tolerated in view of its repetitive nature. MRI examinations are invariably more costly than CT, in view of the longer examination times. However, implementing these techniques in two cohorts of patients: (1) children with benign pathology, (2) patients already undergoing MRI and who would ordinarily require additional CT imaging, offers a cost-effective, efficient approach with radiation protection in the most vulnerable patient group.