MRI-based cephalometrics: a scoping review of current insights and future perspectives

Abstract

Objective:

This review aims to explore the current status of magnetic resonance imaging (MRI) as a cephalometric tool, summarize the equipment design and methods, and propose recommendations for future research.

Methods:

A systematic search was conducted in electronic databases, including PubMed, Ovid MEDLINE, Scopus, Embase, Web of Science, EBSCOhost, LILACS, and Cochrane Library, using broad search terms. The articles published in any language till June 2022 were considered. Cephalometric studies conducted using the MRI dataset on human participants, phantom or cadaver were included. Two independent reviewers assessed the final eligible articles using the quality assessment score (QAS).

Results:

Nine studies were included in the final assessment. Studies used various methods, including 1.5 T or 3 T MRI systems and 3D or 2D MRI datasets. Among the imaging sequences, T ₁-weighted, T ₂-weighted and black bone MR images were used for cephalometric analysis. In addition, the reference standards varied among studies, such as traditional 2D cephalogram, cone-beam CT and phantom measurements. The mean QAS of all the included studies was 79% (± 14.4%). The main limitation of most studies was the small sample size and the heterogeneity of the methods, statistical tools used, and metric outcomes assessed.

Conclusions:

Despite the heterogeneity and lack of metrological evidence on the effectiveness of MRI-based cephalometric analysis, the preliminary results demonstrated by in vivo and in vitro studies are encouraging. However, future studies exploring MRI sequences specific to cephalometric diagnosis are required for wider adoption of this technique in routine orthodontic practice.

Introduction

It is said that cephalometrics is the language in which the poetry of orthodontics is written. ¹ Over the years, cephalometric analyses have been integral to orthodontic diagnosis, treatment planning, evaluation of treatment outcomes, and research tools. The core components of typical orthodontic cephalometry haven't changed much since Holly B. Broadbent introduced cephalometry in the early 1930s. ² The enhanced capabilities of computers and the technological advances in imaging techniques were adopted in orthodontic radiology to develop digital radiography, improve the quality of images, and reduce radiation doses. ³ The drawbacks of film-based two-dimensional (2D) radiographs, such as image magnification, distortion, and other projection errors, were minimised but not wholly eliminated. ²

Three-dimensional (3D) imaging was introduced to overcome the drawbacks of 2D radiographs and meet the demands of advancing technologies in oral diagnosis and treatment planning. The concept of computerised transverse axial scanning paved the way for Ambrose and Hounsfield’s development of computed tomography (CT) in 1973, with the first clinical scan performed at the Mayo Clinic in the United States. ⁴ The reasons for the slow uptake of CT in dentistry were its high cost, limited access, the low resolution of dental structures, and significant radiation exposure. In the late 1990s, Arai in Japan and Mozzo in Italy independently developed Cone Beam Computed Tomography (CBCT) technology for the oral and maxillofacial region with the first commercial machine (NewTom QR-DVT 9000) manufactured by QR Srl of Verona, Italy. ^5,6 The CBCT is a cost-effective and low radiation modality compared to conventional CT, which has sped up its emergence into dentistry. In addition, it has enabled the visualizations of anatomical structures, previously thought impractical. Apart from 3D volumetric and multiplanar reconstruction views, different 2D views can be generated from a 3D dataset. ^7,8

The radiation levels of CBCT are relatively lower than CT but higher than the 2D cephalogram and panoramic radiograph. The effective dose of the CBCT (132–210 μSv) was reported to be 5 and 7 times higher than the combined doses of a panoramic radiograph and lateral cephalogram (26.9–30 μSv). ⁹ Therefore, the large FOV (field of view) CBCT cannot be indicated solely to synthesize 2D cephalograms for routine orthodontic diagnosis and treatment planning. ^10,11 The “Image Gently” campaign was started in response to the concern regarding the harmful effects of ionising radiation, more so on the paediatric population. ¹² Substantial efforts have been made to reduce radiation dose in orthodontic patients, such as introducing ultra-low-dose protocols. ¹³ Similarly, a biplanar low-dose X-ray imaging system has been developed to obtain lateral and frontal 2D cephalograms simultaneously and can be reconstructed into 3D images using mathematical algorithms. ¹⁴ Although there is a possibility for reduced radiation exposure, theoretically, even a small amount raises the lifetime risk from stochastic effects. ¹⁵ Furthermore, radiation hazard from exposure to serial radiographs during the treatment and follow-up is a serious concern in orthodontics, especially in growing children and adolescents. Therefore, it is desired to adopt an imaging modality that can avoid ionizing radiation. ¹²

Magnetic resonance imaging (MRI) is a non-invasive medical imaging modality, an alternative to CT, used widely to evaluate soft tissue pathologies. ¹⁶ Paul C. Lauterbur invented MRI in 1971. It was further refined by Peter Mansfield, for which they were awarded Nobel Prize in Physiology or Medicine 2003. ¹⁷ In orthodontics, conventional MRI systems have been used to evaluate the temporomandibular joint complex, ¹⁸ analyse the upper airway anatomy, ¹⁹ and identify the obstruction sites in obstructive sleep apnoea patients. ²⁰ However, apart from the poor distinction between bone and air, the factors that may limit the use of conventional MRI systems for cephalometric analyses include longer scanning time, complex post-processing methods, difficulty in identification of complex anatomical structures, and metal artefacts from the pre-existing metal restorations, fixed appliances, and fixed retainers in the oral cavity. ²¹ The first attempt to use the MRI modality as a cephalometric tool was the “black bone imaging” reported by Eley et al. in 2013. ²² Due to various drawbacks, Abkai et al developed a novel scanning protocol to obtain an orthogonal cephalometric projection. ²¹ Besides the 2D technique, various studies focussed on performing 3D cephalometric analysis. ^23–27 The validity and accuracy of different methods have been compared with the gold standard and found comparable.

Modern MRI systems employing a high-field scanner, dedicated coil systems, application-optimised sequences, ultrashort echo time, and post-processing free methods may evolve as a promising modality in the near future. ^{21,23,24,27,28} The use of MRI as a cephalometric tool has been explored with the possibility of acquiring a high-resolution isotropic dataset. ²⁸ Similar to CBCT, the multiplanar reconstructive views of the MRI 3D dataset can be used to identify landmarks. (Figure 1) However, while studies on the effectiveness of MRI for cephalometric analysis have shown encouraging results, a wide range of acquisition parameters and different scanning methods have been reported. Therefore, due to limited organised information, we aimed to investigate the current status of MRI as a cephalometric tool, summarise the protocols, and provide a roadmap for future studies.

Figure 1.

Some common skeletal, dental and soft tissue landmarks depicted in the multiplanar reconstructed view (sagittal) of T ₂-weighted MR images (N', soft tissue nasion; Prn, pronasale (nose tip); UL, upper lip; LL, lower lip; Pog', soft tissue pogonion; Me', soft tissue menton; N, nasion; S, sella; Ba, basion; ANS, anterior nasal spine; PNS, posterior nasal spine; Pog, pogonion; Me, menton; T, tip of tongue; P, tip of soft palate; U1T, tip of maxillary central incisor; U1A, apex of maxillary central incisor; U6, mesiobuccal cusp tip of maxillary first molar; L6, mesiobuccal cusp tip of mandibular first molar; Co, condylion; Go, gonion)

Methods and materials

Protocol and registration

A preliminary search of the MEDLINE database, PROSPERO, Open Science Framework, Cochrane Database of Systematic Reviews, and JBI Evidence Synthesis revealed no registered or ongoing review. The current scoping review was registered in the OSF Registries (https://doi.org/10.17605/OSF.IO/JBDUX).This review was performed and reported according to the Preferred Reporting Items for Systematic review and Meta-Analyses with extension for Scoping Reviews (PRISMA-ScR) guidelines ²⁹ in tandem with Joanna Briggs Institute (JBI) guideline for scoping reviews. ³⁰

Eligibility criteria

The inclusion criteria were developed using the “Population–Concept–Context (PCC)” framework recommended by JBI Reviewer’s Manual for Scoping Review. ³⁰

Population: Cephalometric studies conducted using an MRI dataset on human participants with no age restriction, and cadaver or phantom models designed for use in MRI studies were included. Studies that enrolled patients with craniofacial anomalies, those undergoing fixed orthodontic treatment, and those with a history of orthognathic surgery or facial trauma were excluded.

Concept: Articles that reported on the accuracy, validity, or reproducibility of landmark identification and 2D or 3D cephalometric measurements using MRI data were included in this review.

Context: No restrictions were applied based on geographic location, ethnicity, gender, study setting and language.

Information and search

A literature search was conducted using various electronic databases, including PubMed, Ovid MEDLINE, Scopus, Embase, Web of Science, EBSCOhost, LILACS, and Cochrane Library. A broad search was performed using the keywords consisting of “magnetic resonance imaging”, “cephalometrics”, “orthodontics”, “oral surgery”, and “maxillofacial radiology”. The detailed search strategy for each database is given in Table 1. The search included articles published till June 30, 2022. The publication types such as conference abstract, book chapter, personal opinion and letter to the editor were excluded.

Table 1.

Search strategy used in different databases

Database	Search strategy	No. of articles
PubMed	(“Magnetic Resonance Imaging”[MeSH Terms] OR “Magnetic Resonance Imaging” OR MRI OR “T1 weighted” OR “T 1-weighted” OR T1WI OR “T2 weighted” OR “T 2-weighted” OR T2WI OR “black bone”) AND (“cephalometry”[MeSH Terms] OR cephalometry OR cephalometric OR cephalogram) AND (orthodontics[MeSH Terms] OR orthodontic* OR “surgery, oral”[MeSH Terms] OR “oral surgery” OR “oral radiology” OR maxillofacial)	214
Web of science	((TS=(('magnetic resonance imaging' OR mri OR 'T1 weighted' OR 'T 1-weighted' OR T1WI OR 'T2 weighted' OR 'T 2-weighted' OR T2WI OR 'black bone'))) AND TS=((cephalometry OR cephalogram OR cephalometric))) AND TS=((orthodontic* OR 'oral surgery' OR 'oral radiology' OR maxillofacial))	185
Scopus	TITLE-ABS-KEY ((“Magnetic Resonance Imaging” OR mri OR “T1 weighted” OR “T 1-weighted” OR T1WI OR “T2 weighted” OR “T 2-weighted” OR T2WI OR “black bone”) AND (cephalometry OR cephalogram OR cephalometric) AND (orthodontic* OR “oral surgery” OR “oral radiology” OR maxillofacial))	182
Embase	(('magnetic resonance imaging' OR mri OR 'T1 weighted' OR 'T 1-weighted' OR T1WI OR 'T2 weighted' OR 'T 2-weighted' OR T2WI OR 'black bone') AND (cephalometry OR cephalogram OR cephalometric) AND (orthodontic* OR 'oral surgery' OR 'oral radiology' OR maxillofacial)):ti,ab,kw	50
Ovid MEDLINE	(“Magnetic Resonance Imaging” or MRI or “T1 weighted” or “T 1-weighted” or T1WI or “T2 weighted” or “T 2-weighted” or T2WI or “black bone”).ti,ab,kw. and (cephalometry or cephalogram or cephalometric).ti,ab,kw.	73
LILACS	(('magnetic resonance imaging' OR mri OR 'T1 weighted' OR 'T 1-weighted' OR T1WI OR 'T2 weighted' OR 'T 2-weighted' OR T2WI OR 'black bone')) AND ((cephalometry OR cephalogram OR cephalometric)) AND ((orthodontic* OR 'oral surgery' OR 'oral radiology' OR maxillofacial)) (Title, Abstract, Subject)	92
EBSCO	AB (“Magnetic Resonance Imaging” OR mri OR “T1 weighted” OR “T 1-weighted” OR T1WI OR “T2 weighted” OR “T 2-weighted” OR T2WI OR “black bone”) AND AB (cephalometry OR cephalogram OR cephalometric) AND AB (orthodontic* OR OR “oral surgery” OR “oral radiology” OR maxillofacial)	51
Cochrane	((“Magnetic Resonance Imaging” or MRI or “T1 weighted” or “T 1-weighted” or T1WI or “T2 weighted” or “T 2-weighted” or T2WI or “black bone”)):ti,ab,kw AND ((cephalometry or cephalogram or cephalometric)):ti,ab,kw AND ((orthodontic* or “oral surgery” or “oral radiology” or maxillofacial)):ti,ab,kw	8

Selection of sources of evidence

The search results were transferred to Endnote Online software (Clarivate, Philadelphia, Pa) to identify and remove duplicates. The titles and abstracts of the retrieved citations were then independently reviewed by two observers (K.S. and M.S.) to exclude non-eligible articles in accordance with the eligibility criteria. The full text of the studies that were deemed possibly potential for inclusion was examined by the same two observers. A manual search was performed on Google Scholar and grey literature. Only the first 300 records, as sorted in the relevance order of Google Scholar, were included for screening. A manual search of references cited in the included papers was also conducted. Any disagreement regarding the inclusion of the studies was resolved by discussion between the two observers (K.S. and M.S.). However, the senior authors made the final decision if no agreement could be reached.

Data charting process and data items

Two observers (K.S. and K.M.) conducted the data charting process independently. Any disagreement was resolved by discussion between the two observers; if no consensus could be reached, the senior authors made the final decision. The following data were extracted from the final full articles: authors, country of origin, year of publication, study design, study objective, sample characteristics, MRI system and sequence parameters, reference standard, landmarks and measurements, observer details, statistical methods, the main outcome, and conclusion.

Critical appraisal of individual sources of evidence

Since there was no standardised tool available for critical appraisal of studies on MRI cephalometrics, a customised tool named “Quality Assessment Score (QAS)” had previously been reported by Borotikar et al was used in this study. ³¹ The QAS score rated the intrinsic quality of studies based on the evaluation of study design, quality of the methodology, statistical analysis, and results, and also rated metrological evidence. A percentage was calculated from the total score and rounded to the nearest integers. The score of any field that was not applicable was deducted from the maximum score, and the final percentage was calculated. Two observers (K.S. and M.S.) independently reviewed the included studies and rated the QAS. Any discrepancies between the two observers were resolved by discussion with the senior author, and a consensus was reached.

Synthesis of results

The variability in the study design, reference standard, statistical methods, and outcomes reported in the final studies precluded performing a quantitative synthesis. Therefore, we performed a critical mapping of the literature on this topic.

Results

Selection of sources of evidence

A total of 1161 studies were retrieved by searching the following databases: PubMed (n = 214), Web of Science (n = 185), Scopus (n = 182), Embase (n = 50), Ovid MEDLINE (n = 73), EBSCO (n = 51), LILACS (n = 92), Cochrane (n = 8), OpenGrey (n = 6), Google Scholar (n = 300) and ClinicalTrials.gov (n = 0). After removing the duplicates (n = 743) and subsequent screening of the title and abstracts (n = 418), 45 studies were selected for full-text eligibility assessments. Among the 45 studies, 36 were rejected because of non-MRI (n = 5) and non-MRI cephalometric studies (n = 31). Finally, nine studies ^21–28,32 was selected for data synthesis and critical appraisal. No additional studies were found after manually searching the references of the included studies. The PRISMA flow diagram depicting the process of identification, inclusion and exclusion of the studies is shown in Figure 2.

Figure 2.

The PRISMA flow diagram showing the process of identification, inclusion and exclusion of the studies (*search restricted to first 300 records)

Characteristics of the source of evidence

The main characteristics of the included studies are presented in Table 2. The publication year of the included studies ranged between 2013 and 2021, with the majority published in the previous 2 to 3 years. Two articles were published in PLoS ONE, ^21,28 and the remaining were in Dento-maxillo-facial Radiology, ²² Scientific Reports, ²⁴ Progress in Orthodontics, ²³ Clinical Oral Investigations, ²⁶ European Radiology, ²⁵ European Journal of Orthodontics ²⁷ and International Journal of Scientific Research. ³² Six studies originated from Germany ^21,24–28 and one from Italy, ²³ India, ³² and the UK. ²² Except for one study, ²³ all others were prospective.

Table 2.

Characteristics of included studies

Author Year & Country	Study design	Sample	Index test	Reference standard	MRI system	MRI sequence parameters	Landmarks and measurements	Observers	Statistical tool	Findings/ outcome metrics	Conclusion
Eley et al., 201322 UK	In vivo, prospective	Three patients with a mean age of 37 years (range, 21–60 years) and five patients with a mean age of 19 years (range, 16–27 years)	2D MRI cephalogram (Black bone image, T1WI and T2WI of midsagittal plane)	Conventional 2D LCR	1.5T MRI system	TR: 8.6 ms, TE: 4.2 ms, Scan FOV: 24 cm, Phase encode: 256, Frequency encode: 256, Receive bandwidth: 31.25, NEX: 2, ETL: 1	11 landmarks (nine skeletal and two dental); 17 measurements (11 linear and six angular)	One observer repeated the measurements ten times (time interval and blinding not reported)	Coefficient of variation to assess repeatability between measures	Discrepancy between black bone images and LCR was 2.1° ± 1.7° for the angles and 2.8 ± 2.7 mm for the distances Discrepancy between T1W and LCR was 5.0° ± 2.9° for the angles and 3.3 ± 3.2 mm for the distances Among black bone images, the greatest variability occurred with dental landmarks	Black-bone MRI sequence showed comparable accuracy to that of LCR in landmark identification
Heil et al., 201728 Germany	In vitro and in vivo, prospective	Phantom 20 patients with a mean age of 13.95 ± 5.34 years (Range, 8–26 years)	2D MRI cephalogram (Postprocessed)	Conventional 2D LCR and phantom measurements	3T MRI system with a 16-channel multipurpose coil using T1 weighted, isotropic SPACE sequence	TE: 26 ms, TR: 800 ms, bandwidth: 501 Hz/pixel, number of averages: 2, ETL: 63, FOV: 175× 175 mm, acquisition matrix: 256 × 256, voxel size: 0.68× 0.68 mm ×0.68 mm, number of sections: 192, time of acquisition: 6:59 min	18 landmarks (10 skeletal and dental; 10 midsagittal and eight bilateral); 24 measurements (10 linear and 14 angular)	Two independent observers (one radiologist and one orthodontist) analysed twice in 4 weeks interval (blinding not reported)	Intra- and inter- examiner reliability using ICC Bland-Altman analysis to assess agreement between two modalities Equivalence testing (two one-sided test)	Measurements of MRI sequence were equivalent to phantom values Statistical equivalence between MRI and LCR observed for all measurements except interincisal angle. Highest bias (−1.33) and widest 95% limits of agreement (−7.22, 4.56) was observed for interincisal angle Excellent intraobserver and interobservers reliability for both the modalities	Measurements from LCR derived from high-resolution isotropic MRI datasets has high concordance to the corresponding measurements on conventional LCR.
Juerchott et al., 201824 Germany	In vitro and in vivo, prospective	Phantom Three male patients (27 years; 33 years, and 31 years)	T1WI 3D MRI dataset (Patient data in five different head positions and phantom data in three different head positions)	Phantom to validate accuracy	3T MRI system using a 16-channel multipurpose coil with high-resolution T 1-weighted 3D MSVAT-SPACE prototype sequence applied	TE: 5.8 ms, TR: 800 ms, bandwidth: 625 Hz/pixel, number of averages: 1, ETL: 100, FOV: 171× 171 mm, acquisition matrix: 320 × 320, voxel size: 0.53× 0.53 mm ×0.53 mm, number of sections: 256, time of acquisition: 7:01 min	27 landmarks (skeletal and dental); 45 measurements (26 linear and 19 angular)	One radiologist with 5 year experience (repeated measures and blinding not reported)	Equivalence testing (two one-sided test) Bland-Altman analysis to assess level of agreement between MRI phantom measurements and true values One-way ANOVA with Green house -Geisser correction for reproducibility of in-vivo cephalometric measurements	MRI-based phantom measurements showed statistical equivalence and an excellent agreement (bias ranges: −0.090 to 0.044°, −0.220 to 0.241 mm) In vivo cephalometric analysis was highly reproducible without statistical differences for all angles and distances (average ranges: 0.88°/0.87 mm).	Accurate and reproducible 3D cephalometric analysis can be performed using MRI
Jency et al., 201932 India	In vivo, prospective	11 patients (age 18–30 years)	2D MRI cephalogram (Black bone image, T1WI and T2WI of midsagittal plane)	Conventional 2D LCR	1.5T MRI system	TR: 11 ms, TE: 4.20 ms, FOV: 220 mm	Hard tissue and soft tissue landmarks; 18 measurements (12 linear and six angular)	One radiologist repeated measurements ten times (time interval and blinding not reported)	1. Covariance between LCR and MRI images 2.Paired T-test between mean values of LCR and MRI measurements	The ease of landmark identification was difficult on T2 weighted images, but on black bone images, it was comparable to LCR.	Black bone MRI sequence can be an effective non-ionizing imaging modality over conventional methods.
Maspero et al., 201923 Italy	In vivo, retrospective	18 subjects (four male; 14 female) with a mean age of 37.8 ± 10.2 years	T2WI 3D MRI dataset	CBCT (Acquisition parameters- 4 mm slice thickness, 170 × 230 mm FOV, 20 sec scan time, 0.49 × 0.49×0.5 mm voxel size, 120 kVp, and 3–8 mA)	3T MRI system	TR: 2500 ms, TE: 280 ms, NEX: 1, ETL: 65, bandwidth: 255 Hz/pixel, flip angle: 90°, FOV: 240 × 240×180 mm, voxel size: 0.49 × 0.49×0.50 mm, section thickness: 0.49 mm, and time of acquisition: 5:27 min	18 landmarks (10 midsagittal and four lateral points) 24 measurements (11 linear and 13 angular)	Two independent orthodontist performed the analysis twice in 3 weeks interval	1. Intra- and interobserver agreement by ICC Coefficient of variation (CV) to compare the precision of measurements Two-sample t-test to evaluate the differences in the mean CV between CBCT and MRI Bland-Altman analysis to assess the agreement between the two modalities with 95% limits of agreement Paired t-test to compare CBCT and MRI measurements	1.Both CBCT and MRI showed good reliability, with mean intraobserver ICCs of 0.977/0.971 for CBCT and 0.881/ 0.912 for MRI. Average interobserver ICCs: 0.965 for CBCT and 0.833 for MRI. A similar range of agreement between the two modalities with bias range −0.25 to 0.66 mm and −0.41 to 0.54°	Cephalometric measurements using 3T-MRI possess adequate reliability and repeatability, and satisfying agreement with CBCT measurements.
Juerchott et al., 202026 Germany	In vivo, prospective	16 patients (8 males; 8 females) with a mean age of 23.3 ± 7.5 years (range, 14–40 years)	T1WI 3D MRI dataset	None	3T MRI system using a dedicated 15-channel dental surface coil and T1-weighted 3D MSVAT-SPACE prototype sequence applied	TE: 5.8 ms, TR: 800 ms, bandwidth: 625 Hz/pixel, number of averages: 1, ETL: 100, FOV: 171× 171 mm, acquisition matrix: 320 × 320, voxel size: 0.53× 0.53 mm ×0.53 mm, number of sections: 256, time of acquisition: 7:01 min	42 landmarks [28 skeletal (14 midsagittal and 14 bilateral) and 16 dental)]	Two independent radiologists identified landmarks twice after more than 4 weeks interval (Blinding done)	Intra- and interrater agreement by mean measurement errors and ICC Measurement error calculated as Euclidean distances	Intra- and interrater ICCs were consistently higher than 0.9. Intrarater comparisons showed mean measurement differences (ranges) of 0.87 mm (0.41 to 1.63) for rater I and 0.94 mm (0.49 to 1.28) for rater II. Average interrater difference was 1.10 mm (0.52 to 2.29).	High-resolution 3D MRI enables reliable determination of 3D cephalometric landmarks with high intra- and interrater reliability
Juerchott et al., 202025 Germany	In vivo, prospective	12 patients (8 males, 4 females) with a mean of 26.1 ± 6.6 years (range, 17–40 years)	T1WI 3D MRI dataset	CBCT (Acquisition parameters- tube voltage: 98 kV, tube current: 5 mAs, scanning time: 14 s, FOV: 150 × 150 mm, and isotopic voxel size: 0.25 mm)	3T MRI system using a dedicated 15-channel dental surface coil and T 1-weighted 3D MSVAT-SPACE prototype sequence applied	TE: 5.8 ms, TR: 800 ms, bandwidth: 625 Hz/pixel, number of averages: 1, ETL: 100, FOV: 171× 171 mm, acquisition matrix: 320 × 320, voxel size: 0.53× 0.53 mm ×0.53 mm, number of sections: 256, time of acquisition: 7:01 min	27 landmarks (skeletal and dental) 35 measurements (18 linear and 17 angular)	Two independent radiologists performed the analysis twice after more than 4 weeks interval (Blinding done)	Calculation of Euclidean distances, ICC, Bland- Altman analysis, and equivalence testing (linear mixed effects model) with a predefined equivalence margin of ±1°/1 mm.	Analysis of reliability for CBCT vs MRI (intrarater I/ intrarater II/ interrater) revealed Euclidean distances of 0.86/0.86/0.98 mm vs 0.93/0.99/1.10 mm for landmarks, ICCs of 0.990/0.980/0.986 vs 0.982/0.978/0.980 for angles, and ICCs of 0.992/0.988/0.989 vs 0.991/0.985/0.988 for distances. High levels of agreement with bias value of 0.03° (− 1.49 to 1.54) for angles and 0.02 mm (− 1.44 to 1.47) for distances. The mean values of CBCT and MRI measurements were equivalent.	3T MRI enables reliable 3D cephalometric analysis and excellent agreement with CBCT measurements
Marz et al., 202127 Germany	In vitro, prospective	Three human cadaver head preparations	T1WI 3D MRI dataset with no post-processing	Conventional 2D LCR	3T MR scanner and a standard 20-channel head-neck-coil was used and T1-weighted multislice Turbo Spin Echo MR images was obtained	TE: 9.9 ms, TR: 300 ms, turbo factor: 3, in-plane spatial resolution: 0.88 × 0.88 mm2, slice thickness:1.5 mm, 120 slices, total acquisition time: 19.5 min, three averages	19 skeletal and dental landmarks; 13 angular measurements	Five independent orthodontists (Blinding done and repetitive measures not reported)	ICC for agreement between the five raters Bland-Altman plots to visualise interrater agreement Linear mixed-effects model for angle specific differences and equivalence testing (two one-sided test) including Bonferroni-Holm correction	The interrater reliability was high for all angles (ICC≥0.74) Differences between LCR and MRI measurements ranged between –0.91° (–1.88 to 0.07) and 0.97° (–0.63 to 2.57) and were equivalent with respect to a margin of [–2°, 2°] except for lower incisor inclination.	A reliable method for cephalometric analysis of a 3D MRI dataset with semi-automatic dataset orientation was established
Abkai C et al., 202121 Germany	In vitro, prospective	One patient (male)	7 MRI cephalometric projections (MCPs) with various scan parameters and no post-processing	Conventional 2D LCR	3T MRI system with eight-channel head coil	Pixel size 0.39 × 0.39 mm or 0.2 × 0.2 mm, FOV 293.3 × 293.3 mm or 300 × 300 mm, scan time ranging from 5 to 154 sec, TE: 358 to 360 μs, TR: 4.2 to 16.9 ms, pixel bandwidth: 816, 517 or 259 Hz and NEX: one or 2	14 skeletal and dental landmarks; 10 angular measurements	40 orthodontists with at least 15 years' experience (randomisation of radiographs and blinding done;repetitive measures not reported)	Levene’s test for evaluating homogeneity of variance and two-tailed t-tests	Mean relative distances were 2.4 to 2.7 mm in MCPs and 1.6 mm in LCR, demonstrating the accuracy and level of agreement of assessors MCP showed on average deviation of 1.2˚ ±1.1˚ in comparison to LCR. Larger differences were found for gonial and interincisal angles. Images with higher resolution and contrast (154 s) showed mean angular deviations up to 0.9˚ and lower resolution images (5–16 s) showed mean angular deviation up to 1.6˚ in comparison to LCR.	MCPs can be acquired much faster, and this study demonstrated the potential of this new method.

ANOVA, Analysis of Variance; CBCT, Cone Beam Computed Tomography; 2D, two-dimensional; 3D, Three dimensional; ETL, Echo train length; FH, Frankfort horizontal; FOV, Field of View; Hz, Hertz; ICC, Intraclass Coefficient; LCR, Lateral Cephalometric Radiograph; MRI, Magnetic Resonance Imaging; MSVAT-SPACE, Multiple-Slab acquisition with View Angle Tilting gradient based on Sampling Perfection with Application optimised Contrast using different flip angle Evolution; NEX, Number of excitations; SPACE, Sampling Perfection with Application optimised Contrast using different flip angle Evolution; T, Tesla; TE, echo time; TR, repetition time; T1WI, T1 Weighted Images; T2WI, T2 Weighted Images; cm, centimetres; kV, tube voltage; kVp, kilovoltage Peak; mA, milliampere; min, minutes; mm, millimetres; ms, milliseconds; s, second; μs, microseconds.

Critical appraisal within sources of evidence

The results of the QAS of included studies are presented in Table 3. The mean QAS of included studies was 79% (± 14.4%). Two articles had a QAS of 95%, and both reported validity and intraobserver and interobserver reliability of the MRI technique. ^25,28 Two studies had a QAS between 80 and 90%. ^23,26 Three studies had a QAS ranging from 70 to 80%, ^21,24,27 and the remaining two had a QAS of 62 and 55%. ^22,32 Based on the QAS, most articles showed good intrinsic quality in reporting objectives, study design, sample description, index test and reference standards, equipment design, acquisition parameters, outcome measures, statistical analyses, and study limitations. However, the main drawback was that none of the studies reported sample size calculation. In addition, most studies reported the details of observers/ raters; however, only four studies reported avoidance of test-retest bias such as blinding and time interval. ^21,25–27

Table 3.

The Quality Assessment Scores of included studies

Items	Key for scoring	Eley et al., 2013 22	Heil et al., 2017 28	Juerchott et al., 2018 24	Jency et al., 2019 32	Maspero et al., 2019 23	Juerchott et al., 2020 26	Juerchott et al., 2020 25	Marz et al., 2021 27	Abkai et al., 2021 21
Are the aims of the study clearly stated?	Clear (2)/ Partial (1)/ No (0)	1	2	2	1	2	2	2	2	2
Is the study design prospective or retrospective?	Prospective (2)/ Both (1)/retrospective or not reported (0)	2	2	2	2	0	2	2	2	2
Is there an adequate description of the patients or models?	Clear (2)/ Partial (1)/ No (0)	2	2	2	1	2	2	2	2	2
Are details about inclusion and exclusion criteria provided?	Clear (2)/ Partial (1)/ No (0)	2	2	1	2	2	2	2	2	2
Was volunteer/patient consent obtained before the study?	Stated (2)/ Not stated (0)	2	2	2	2	2	2	2	NA	2
Are details about sample size calculation provided?	Yes (2)/ Partial (1)/ No (0)	0	0	0	0	0	0	0	0	0
Is there a clear description of equipment design and set-up of index test?	Clear (2)/ Partial (1)/ No (0)	2	2	2	1	2	2	2	2	2
Are the details of type of acquisition and acquisition parameters of index test provided?	Clear (2)/ Partial (1)/ No (0)	2	2	2	1	2	2	2	2	2
Are the details of equipment and acquisition parameter of reference standard provided?	Clear (2)/ Partial (1)/ No (0)	0	2	2	0	2	NA	2	2	2
Is the description of observer/reviewer/rater provided?	Clear (2)/ Partial (1)/ No (0)	1	2	2	1	2	2	2	2	2
Are the main outcomes of the study clearly stated?	Clear (2)/ Partial (1)/ No (0)	2	2	2	2	2	2	2	2	2
Is there a clear description of the measures?	Clear (2)/ Partial (1)/ No (0)	2	2	2	2	2	2	2	2	2
Is there a clear statement of statistical analysis or validity measures conducted?	Clear (2)/ Partial (1)/ No (0)	1	2	2	1	2	2	2	2	2
Are the key findings supported by the results?	Yes (2)/ Partial (1)/ No (0)	2	2	2	2	2	2	2	2	2
Are the key findings answered to the initial objectives?	Yes (2)/ Partial (1)/ No (0)	2	2	2	2	2	2	2	2	2
Is there a description of study limitations?	Clear (2)/ Partial (1)/ No (0)	1	2	1	2	1	2	2	2	2
Is concurrent validity evaluated?	Yes (4)/ Partial (2)/ No (0)	2	4	4	2	4	0	4	4	4
Is interobserver reliability evaluated?	Yes (4)/ Without quantification/clinical relevance (2)/ no (0)	0	4	0	0	4	4	4	4	0
Is intraobserver reliability evaluated? or Was intrasubject reliability evaluated?	Yes (4)/ Without quantification or clinical relevance (2)/ No (0)	2	4	0	0	4	4	4	0	0
Are the criteria for the avoidance of test-retest bias specified?	Yes (4)/ Partial information (2)/ No (0)	2	4	NA	NA	4	4	4	0	NA
Subtotal score (1-16)		24	30	28	22	27	28	30	28	30
Subtotal score (17-20)		6	16	4	2	16	12	16	8	4
Total score		30	46	32	24	43	40	46	36	34
Total QAS (%)		62	96	73	55	89	87	96	78	77

NA, not applicable; QAS, Quality Assessment Scores

Results of individual sources of evidence

Sample

The sample size varied from 3 to 20 individuals, except for a study ²¹ that acquired seven cephalometric projections from a single patient to compare different acquisition parameters. Many studies have recruited both males and females. Both adults and adolescents, ranging in age from 8 to 60 years old, were included in the studies. The inclusion and exclusion criteria were reported in all the studies. Along with in vivo reproducibility, two studies tested the geometric accuracy of the MRI sequence using the phantom models. ^24,28 Three cadaver heads were used as the sample in another investigation. ²⁷ Adequate information was provided on cadaver head processing and position during exposure to MRI and 2D radiography.

Index test

The MRI cephalometric technique used both two-dimensional and three-dimensional datasets. The 3D dataset was acquired in five studies, four using a 3D T ₁-weighted sequence ^24–27 and one using a 3D T ₂-weighted sequence. ²³ The remaining four studies used a 2D MRI dataset. ^21,22,28,32 Among them, one study performed postprocessing. ²⁸ Two studies used black bone MRI sequence, T ₁-weighted and T ₂-weighted images, of the midsagittal plane. ^22,32 One study used multiple cephalometric projections acquired using different scanning protocols. ²¹ Most studies used 3T MRI systems, and only two reported using 1.5 T MRI systems. ^22,32 Essential scanning protocols have been elaborated on in all the studies. The detailed MRI sequence parameters are provided in Table 2.

Reference standard

Among the included studies, eight were comparative; four studies compared MRI with 2D conventional lateral cephalograms (LCR), ^21,22,27,32 two studies compared with CBCT, ^23,25 one study used both phantom and 2D LCR, ²⁸ and the remaining study used the phantom model only to validate the accuracy. ²⁴ Among five studies, three reported the exposure parameters of 2D LCR; tube output of 72kV/15mA, ²⁸ 73kV/16mA ²⁷ and 77kV/14mA, ²¹ source to midsagittal plane distances of 1.5 m ²⁸ and 1.7 m ²⁷ with exposure time of 9.4 s ^27,28 and 9.2 s. ²¹ Among the two studies that used CBCT, the exposure parameters varied: tube voltage 98kV ²⁵ and 120kV, ²³ tube current 5mA ²⁵ and 3-8mA, ²³ the scan time of 14 s ²⁵ and 20 s, ²³ and different FOV, scanning time and voxel size. Among the two studies that used phantom, Heil et al used an American College of Radiology (ACR) accredited MRI phantom to test the geometry accuracy of the MRI technique. ²⁸ In contrast, Juerchott et al used a custom-made cuboid-shaped phantom created from LEGO bricks. ²⁴ Before scanning, the phantom was placed in a plastic box filled with water and a contrast agent.

Measurements and statistical analysis

Eight studies used only hard tissue (skeletal and dental) landmarks, ^21–28 whereas the remaining study used hard tissue and soft tissue landmarks. ³² Both midsagittal and bilateral landmarks have been used in most of the studies. Six studies performed linear and angular measurements, ^{22–25,28,32} whereas two performed only angular measurements, ^17,23 and one performed only landmark identification. ²⁵ All the measures in 2D LCR were computed using software; two studies used Dolphin imaging software, ^22,32 the remaining used OnyxCeph ²⁷ and Romexis software, ²⁸ and one used a web-server-based customised software. ²¹ Most of the studies that used conventional 2D LCR accounted for the magnification errors through calibration tools in software applications. The studies performed 3D cephalometric landmark identification and measurements using Osirix ^24–26 and Mimics software. ²³ Three studies employed one observer, ^22,24,32 four studies had two independent observers, ^23,25,26,28 one reported using five observers, ²⁷ and another had 40 observers. ²¹ All the studies mentioned the field and experience of the observer. Repetitive measures were performed in six studies ^{22,23,25,26,28,32} ; however, two studies did not report time intervals. ^22,32 Blinding of measurements/ radiographs was reported only in three studies. ^25–27

Covariance and coefficient of variation were calculated in three studies to evaluate the reliability, ^22,23,32 and the level of agreement was calculated using the Bland–Altman plot in five studies. ^{23–25,27,28} Intra- and/or interobserver agreement was computed using the Intraclass correlation Coefficient (ICC) test. Equivalence testing between two modalities was performed using two one-sided tests in three studies, ^25,27,28 paired t-test in two studies ^23,32 and a two-tailed t-test in two studies. ^21,23 Measurement errors were calculated based on Euclidean distances in two studies. ^25,26 One-way ANOVA with Greenhouse-Geisser correction was used to evaluate reproducibility in one study, ²⁴ and another used a linear mixed-effects model for angle-specific differences. ²⁵

Outcome assessed and the findings

The main goals of most of the investigations were to evaluate the accuracy and reliability of 3D MRI data or reconstructed 2D MRI images for use in cephalometric analysis. Concurrent validity was assessed in six studies, ^{21,23–25,27,28} and intraobserver and interobserver reliability in five studies. ^23,25–28 Two studies that evaluated the validity of Black bone images with 2D LCR showed that the accuracy of landmark identification and measurements were comparable and superior to the other routine MRI sequences. ^22,32 The discrepancy between “Black Bone” and LCR was 2.1° ± 1.7° for the angles and 2.8 ± 2.7 mm for the distances. However, QAS were very low for these two studies (62 and 54%). The other three studies that compared MRI and 2D LCR showed a statistical equivalence for all measurements except those that used dental landmarks such as interincisal and lower incisor inclination angles. ^21,27,28

Two studies that used CBCT showed a similar level of agreement between CBCT and MRI measurements, and the equivalence testing was also significant between them. ^23,25 Similarly, the two studies that compared the measurements of MRI sequences with known phantom values showed a high concordance proving the validity of MRI measurements. ^24,28 Studies that reported intraobserver and interobserver reliability showed that the MRI measurements possess adequate reliability and repeatability except for the measurements related to the dental landmarks and gonial angle (due to articulare landmark). The ICC for intra- and interobserver ratings of the majority of studies were consistently above 0.9. A recent study that compared the reliability of 40 observers showed that mean relative distances were 2.4–2.7 mm in MRI cephalometric projection and 1.6 mm in LCR, demonstrating the accuracy and level of agreement of assessors. ²¹ The detailed findings of the studies are provided in Table 2.

Discussion

This review attempted to identify the methods of MRI-based cephalometric analysis and report the current status. Nine studies were identified on this topic and reviewed. Among them, only four studies had a QAS of above 80% (Table 3). ^23–25,28 The main limitation of most studies is the small sample size that ranges from one to twenty individuals. Furthermore, the lack of an a priori sample size estimate also makes it difficult to assess the quality of the metric results and ensure the adequate power of the study. Another drawback was the heterogeneity of the statistical tool and metric outcomes assessed, which made it difficult to use a standard scale for comparison and reporting. It is essential to report the standard error of measurement, which is a function of both the reliability and the standard deviation, so that the observed difference may be attributed to either a true measurement of change or a measurement error.

Traditionally, MRI scans are used to evaluate soft tissue pathologies and variations because of their high water/proton content. Whereas bone has a low proton density and a very short T2 relaxation time, making it difficult to differentiate between air and bone in conventional MRI scans. ³³ Eley et al validated a prototype MRI sequence called the “Black Bone” images that used a short echo time and repetition time with a low flip angle for detail. ³⁴ This caused the reduction of soft tissue contrast and enabled the augmentation of hard and soft tissue boundaries. However, only a single midsagittal slice was obtained to perform the analysis; therefore, the black bone MR images did not find much clinical application as a cephalometric tool. ²² Heil et al developed a short scanning time MRI system and post-processing algorithm that transformed the isotropic T ₁-weighted 3D MRI datasets into 2D cephalograms with midsagittal and bilateral landmarks. ²⁸ However, it was time-consuming and specific postprocessing software and expertise were required. In contrast, Maspero et al used a 3T- MRI system with T₂ weighted sequence that showed a better contrast between hard and soft tissues, high spatial resolution, and short scanning time (less than 6 min). ²³ Recently, a novel scan protocol was devised where an orthogonal MRI cephalometric projection (MCP) was obtained in one shot without needing post-processing or cropping. ²¹ The potential application of this new technique needs to be validated using a large sample size.

Additionally, there have been efforts to use the 3D dataset to perform cephalometric analysis. Juerchott et al developed a prototype sequence, 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), which was explicitly optimized for craniofacial MRI. ²⁴ However, its cost precluded the immediate adoption for routine use. Furthermore, since the current cephalometric norms are based on traditional 2D cephalograms, which might suffer from projection errors, the direct comparison of cephalometric measurements obtained from a 3D MRI dataset with the traditional cephalometric norms may not be accurate. ^35,36 Furthermore, no comprehensive 3D cephalometric norms exist currently due to ethical issues with subjecting healthy individuals to CBCT or CT. Therefore, the development of 3D cephalometric norms using 3D MRI datasets can revolutionise the use of 3D cephalometric analysis in clinical practice and become a suitable alternative to 2D LCR or CBCT.

Head positioning is critical during the cephalometric examination, particularly when serial radiographs are to be compared. ³⁷ In contrast to traditional cephalometric equipment or CBCT machines, the patient is positioned on the table, usually supine, during MRI acquisition. Studies have shown significant differences in the soft tissue around the chin and neck when the patient’s head position is supine, which might influence orthodontic and orthognathic surgical planning. Additionally, a few airway-related parameters have also been shown to alter. ³⁸ These differences may further limit the direct comparison of measurements obtained from traditional LCR, CBCT/CT, and MRI images. However, with open MRI systems, where the patient can be made to sit or stand, the error pertaining to patient positioning can be eliminated. The motion artefact associated with claustrophobic fear and the need for sedation can be avoided. ³⁹ Moreover, they are low-field MRI scanners that may have compromised image quality and require longer scanning time. ⁴⁰ The application of open MRI systems for cephalometric purposes needs to be explored in future studies.

Another primary concern with MRI acquisition is the presence of orthodontic appliances or fixed lingual retainer, which may be ferromagnetic and cause metal artefacts. ^41,42 Additionally, patients with osteosynthesis materials or dental restorations are also susceptible to metal artefacts. ^43,44 Studies have shown that more significant artefacts are seen in 3T MRI compared to 1.5T MRI, stainless steel material showed more distortion than titanium and ceramic materials, and a gradient-echo sequence (GRE) increased artefacts compared to a turbo-spin-echo (TSE) sequence. ⁴³ Technical advances in dental MRI systems, including high-field imaging and optimized sequence such as Zero TE (ZTE), have demonstrated rapid imaging times, silent scanning, high-resolution isotropic datasets, and resistance to artefacts. ⁴⁵ Similarly, the motion artefacts brought on by claustrophobia, loud noises, or involuntary motions such as swallowing or breathing are also critical during MRI acquisition.

To reduce the motion and susceptibility artefacts, various MRI sequences were developed, such as zero TE (ZTE), view angle tilting (VAT), slice-encoding for metal artefact correction (SEMAC), multiacquisition with variable resonance image combination (MAVRIC), periodically rotated overlapping parallel lines with enhanced reconstruction (PROPELLER) with fast-spin echo (FSE) and combination of these. ^45–47 However, the role of these sequences in the cephalometric analysis is yet to be explored. Automated craniofacial 3D reconstruction from MRI datasets has been attempted to improve the clinical utility of the MRI. This segmentation is further enhanced by combining various algorithms like ZTE, gradient echo (GRE), black bone (BB) techniques and FIESTA-C (fast imaging employing steady-state acquisition). ⁴⁸ In the near future, it might be appealing to use Artificial Intelligence based MRI cephalometric reconstructions and analysis, which can prove to be an accurate and time-efficient approach. ^49,50

Limitations

It is noteworthy that the included studies are limited due to the novelty of this technique and the cost associated with developing MRI sequences specifically for cephalometric analysis. A systematic review with quantitative analysis will be conceivable if more studies comparing different MRI systems/sequences using a similar reference standard and reporting similar outcome measures are available. Another limitation of this review is the use of the QAS system adopted from the previous study. Even although the QAS is based on well-known and validated quality assessment tools, the validity of QAS is unknown. Therefore, care must be taken when interpreting the QAS reported in this study.

Conclusion

The MRI-based cephalometric approach is in a nascent stage of development. Currently, there is a lack of metrological evidence for their use as a tool for cephalometric analysis. Nevertheless, the results demonstrated by in vivo and in vitro studies are encouraging. Furthermore, using non-radiation techniques can prove beneficial in orthodontics, especially in young patients and instances where radiation exposure may be a concern. However, the processing technique, cost and equipment design could limit the immediate adoption of this modality for routine use in cephalometric analysis.