Establishment of the mid-sagittal reference plane for three-dimensional assessment of facial asymmetry: a systematic review

The systematic review reported follows the PRISMA guidelines: Preferred Reporting Items for Systematic Reviews and Meta-Analyses [18]. The review protocol was registered with the PROSPERO database (registration number: CRD42020218963; https://​www.​crd.​york.​ac.​uk/​prospero/​display_​record.​php?​ID=​CRD42020218963).

The following focused topic that matched the Population-Intervention-Control-Outcome (PICO) criteria was the subject of a literature search: “What is the best technique to construct a mid-sagittal reference plane for the estimation of facial asymmetry?” For this review, studies that addressed the aforementioned question were judged appropriate. The elements for PICO criteria have been listed in Table 1. Studies were considered if they met the following inclusion criteria: 1) utilized a 3D technology-based tool, device, or software or conducted any intervention for the estimation of MSP; 2) conducted on humans and had the proper analytical design, such as case–control studies, cross-sectional studies, prospective studies, and retrospective studies, including pilot studies; 3) study data did not duplicate or overlap with those of other articles; 4) studies had full text availability and were published in English. Studies performed on animals, non-human models, non-pertinent data, not involving facial asymmetry subjects or focused on 2D analysis, letters to editors, conference papers, and review articles were excluded from the current systematic review.

All the relevant studies were identified systematically and independently by two authors (DA and PS) through a comprehensive search in the electronic databases:PubMed, EMBASE (via Ovid), Medline (via Ovid), Cochrane Library, Scopus, and Web of Science (until June 2023) with the combination of Medical Subject Headings (MeSH) terms as keywords. Moreover, vocabulary and syntax were adjusted across the databases. The literature search was not constrained by publication date or status. In addition, manual and OpenGrey database (http://​www.​opengrey.​eu/​) searches were carried out.

Data extraction was performed independently by the two reviewers (DA and PS) according to the standardized and predefined data format, recording the following outcomes: 1) effectiveness of the technique; and 2) ease of clinical applicability. Accordingly, the following data were extracted from the full text articles:

The methodological quality of each paper for the risk of bias and applicability was assessed using customized assessment criteria based on the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) [19] (Supplementary Appendix 1). Two reviewers (DA and PS) rated each study independently and assigned a score of low, high, or unclear based on the information presented in the study.

The strength of the evidence for the included articles was determined using the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) scale [20]. The level of evidence was categorized as high, moderate, low, or very low. The ratings were downgraded for the studies exhibiting serious or very serious concerns pertaining to publication bias, inconsistency, imprecision, indirectness, or risk of bias.

The PRISMA flowchart presented in Fig. 1 shows the study selection process. Initially, 6746 records were identified through a comprehensive search across six databases, and 8 records were identified from additional sources. After excluding 66 duplicates, the titles and abstracts of 6680 articles were screened. Of those, 6640 articles were excluded due to their irrelevance to the topic. Following initial screening, a total of 48 articles (40 from the database search and 8 from additional sources) were sought for full text retrieval; however, only 47 potentially eligible articles were assessed for full text review as the full text for one article was unavailable. Following a detailed review of the full text articles, an additional 5 studies were eliminated, and finally, 42 studies that met the inclusion criteria were considered suitable for qualitative analysis. A list of the excluded full text articles along with their justifications is presented in Supplementary Appendix 2. There was excellent inter-reviewer agreement for the study selection process, with Cohen’s κ values of 0.89.

The demographic and methodological characteristics of all the included studies are summarized in Tables 2 and 3. Between 2005 and 2023, a total of 42 studies were identified, focusing on estimating the mid-sagittal plane in patients with facial asymmetry. Out of the selected 42 studies, eight were computational studies [1, 3, 21‐26]; six were observational [14, 15, 27‐30]; 21 studies used retrospective analytical design [4, 9, 9, 12, 31‐47]; three were experimental studies [17, 48, 49], and one study was ambispective in nature [50]. The age range of the patients in all the included studies was 7–57 years. Further, the included studies revealed diverse asymmetry criteria, such as a deviation of 1 mm to 4 mm in soft tissue chin position based on menton (Me), pogonion (Pog), or Gonion (Gn) [4, 12, 15, 24, 28, 30, 32, 9, 33‐35, 38‐45, 47‐53].

The results for the methodological quality and risk of bias assessment have been presented in Fig. 2. All the included studies were representative of the target population; nevertheless, owing to the retrospective, observational, experimental, or computational nature of the included studies, concerns regarding the risk of bias were rated to be relatively ‘high’ in the subject/model selection domain. Within the index test domain, seven studies failed to explicitly define facial asymmetry [3, 16, 17, 22, 31, 9, 46] and four studies inadequately described the process for MSP construction [16, 21, 9, 33]. As a result, these studies were considered to have a ‘high’ risk of bias. However, the overall risk of bias in the index test domain was assessed as ‘low’. Concerning the risk of bias in the reference standard domain, most of the studies were considered to have a ‘low’ risk. With regard to the risk of bias in the domain of workflow, the validity and reliability of MSP construction were questionable in 13 of the included studies [9, 16, 17, 21‐26, 32, 41, 46, 52], and therefore considered to have an ‘unclear/high’ risk of concern. In addition, 19 of the included studies [4, 15‐17, 22‐25, 28, 29, 31, 9, 33, 39, 43, 44, 49, 50] either did not report the specific skeletal discrepancy or focused solely on a particular type, such as Class III malocclusion, juvenile idiopathic arthritis, or craniofacial dysplasia. As a result, the MSP construction methodology described in those 19 studies may not be applicable to all types of facial asymmetry conditions and therefore, they were rated as having a ‘high’ risk of bias. In general, concerns regarding the risk of bias in the domain of workflow were relatively ‘low’.

Concerns about applicability were deemed to be ‘unclear’ in the subject/model selection domain since 19 of the included studies [4, 14, 15, 22, 23, 9, 24, 25, 28, 32‐34, 36‐39, 41, 46, 49, 52] lacked a detailed description of the sampling unit, including the type of skeletal discrepancy, asymmetry criteria, or the patient’s age and gender. The reproducibility of the constructed MSP was found to be questionable in 13 of the included studies [16, 17, 21‐26, 32, 9, 41, 46, 52] as they failed to report the reliability assessment, which led to a rating of ‘high’ risk of concern; however, the overall risk was judged to be ‘low’ in the index test domain. Regarding the reference standard domain, nine studies with ambiguous reference standards were considered to have ‘unclear’ risk [17, 26, 31, 38, 39, 41, 49, 50, 52], while the majority of the studies were deemed to have ‘low’ risk. The Supplementary Figure includes a detailed presentation of the study quality and risk of bias assessment.

Table 4 provides the evidence profile for the outcomes examined in this study. The evidence quality was ‘low’ for the effectiveness of the technique and ‘moderate’ for the ease of clinical applicability.

The majority of the included studies [12, 14, 27‐31, 9, 32, 36, 37, 39‐49, 51‐53] constructed a cephalometric MSP, while eight studies computed a morphometric MSP [3, 17, 21‐24, 34, 38] for the assessment of asymmetry (Table 3). In addition, five studies employed both cephalometric and morphometric MSP construction methods [4, 15, 25, 35, 50], while four studies assessed asymmetry using a symmetry plane [1, 26, 9, 33]. For the cephalometric MSP construction, N-S was the most consistently used landmark combination [12, 28, 35, 39, 40, 43, 47, 48]. Whilst morphometric MSP was determined using modified Procrustes Analysis [35], and Global Registration [34, 38]. The use of the morphometric MSP was found to be accurate and suitable for the evaluation of facial asymmetry in six studies [3, 4, 15, 34, 35, 38]. In particular, based morphometric MSPs showed extremely stable and reliable results (ICC > 0.9) for the estimation of facial asymmetry.

A semi-automatic technique that involves digital landmarking was employed for cephalometric and morphometric MSP construction in nine studies [23, 24, 28, 31, 32, 40, 47‐49], whereas a fully automated and landmark-independent technique for morphometric MSP and symmetry plane construction was reported in seven studies [3, 16, 21, 25, 9, 33, 50]. The fully automated MSP construction method was reported to accurately measure the degree of asymmetry and achieve satisfactory skeletal symmetry, as asserted by the authors in their respective studies (Table 3).

The cephalometric method has been evidenced to be reliable, highly familiar, and a simple technique, as head orientation has no influence on the landmark placement and is therefore convenient in the clinical setting [4, 28]. The present review demonstrated that cephalometric MSP based on N-ANS-PNS was consistent with the symmetric mid-sagittal reference plane [4]. However, a plane passing through the stable landmarks N-S [12] and perpendicular to the horizontal reference plane would be a more appropriate approach if damage or deviation of the ANS were suspected. Ajmera et.al, in their study, utilized N-S based cephalometric MSP and reported it to be equally effective as a clinically derived midline for facial asymmetry estimation [35]. Despite its several advantages, the cephalometric method may seem challenging due to the inherent limitations associated with its landmark digitization step. For instance, midline structures may tip from the true plane of symmetry; stable landmarks need to be identified that are unaffected by asymmetry; errors associated with manual or digital landmark annotation; and the reproducibility of the identified landmarks [4, 15]. The morphometric technique, on the other hand, was developed to overcome some of the shortcomings of the cephalometric method [15]. The strength of the morphometric method is its capacity to use the external face characteristics, which serve as a framework of perceptiveness for asymmetry, to ascertain the true plane of symmetry [15, 57]. Damstra et al. compared cephalometric and morphometric MSPs and suggested using a morphometric approach based on intact regions of the skull, unaffected by asymmetry, to determine MSP [15]. Likewise, Hsiao et al. reported voxel-based morphometric MSP as a more reliable approach compared to the cephalometric method, due to the latter’s dependence on the operator’s skill in identifying landmarks, whereas the former is less prone to deviation even in cases of severe asymmetry [25]. In contrast, a recent study by Ajmera et al. found cephalometric and morphometric approaches (clinically derived midline and modified Procrustes analysis) to be equally effective [35]. The authors further reported that the modified Procrustes method is a practical alternative to conventional Procrustes analysis for evaluating asymmetry. Conventional Procrustes analysis considers all landmarks to achieve an optimal fit, whereas the modified Procrustes method only utilizes four stable landmarks in the upper facial region (bilateral porions and orbitales) that are minimally influenced by asymmetry to achieve the best fit. This was in agreement with Shin et al.’s findings, who reported analogous results for asymmetry measurements (p > 0.05) using morphometric and cephalometric MSP construction approaches [4]. Zhu et al. in their study, compared three different algorithms for morphometric MSP construction and reported that weighted PA-based MSP was a promising approach for cases of severe mandibular deviation [24]. Assessing the various morphometric approaches previously employed, Global registration [34, 38] and modified PA [35] methods were found to be more robust approaches for asymmetry estimation. This is because the Global registration method utilizes a stable anterior cranial base for registration, preventing the influence of other asymmetric regions of the skull, whereas modified PA utilizes four stable landmarks (bilateral porion and orbitale) that are least affected by asymmetry to achieve the ‘best fit’. Interestingly, PA, which has been used in numerous prior investigations, was found to exhibit a masking effect that reduces asymmetry characteristics and may not accurately reflect the true clinical situation [35].

The reliability of the approach is critical for the clinical effectiveness of the MSP construction technique. In the present review, the reliability of the technique was critically analyzed, which revealed that several studies [1, 9, 17, 21‐24, 32] failed to report the reliability assessment despite concluding with encouraging results. For instance, Wong et al. [16, 9, 33] advocated a novel automated and landmark-independent method of MSP construction referred to as the “Symmetry Plane”. Nevertheless, the effectiveness of their technique can be deemed questionable because the information regarding the software used was only briefly described and a reliability assessment was not performed.

For routine clinical use, MSP construction needs to be simple, user-friendly, and ideally fully automatic. Current landmark-dependent methods are simple and user-friendly, but at the same time, they are also technique-sensitive, as they rely on manual or automatic landmarking. While manual digitization is subjective and time-consuming, automatic landmarking may not locate the landmarks precisely and may introduce further errors during asymmetry evaluation. When employing a morphometric approach, the cost of the software and hardware and the need for additional training should not be overlooked [4, 15, 58]. Although the morphometric approach may be quicker to implement, it may not be cost-effective. On the other hand, as the technique is less landmark-dependent and more reliable, it may improve diagnosis and reduce treatment planning time. In this regard, voxel-based morphometric MSP [25] could be a viable option for asymmetry assessment. Another fully automated approach, based on a ‘Symmetry plane’ [9, 16, 33] although showed promising results for routine clinical use, may not be applicable in a routine clinical context owing to the inaccessibility of the in-house developed software.

Despite a comprehensive search and selection of specific studies, this review was limited due to the methodological heterogeneity observed across the included studies, which precluded performing a meta-analysis. Additionally, a lack of standardized assessment of facial asymmetry was noted, which may have had an impact on the findings. In addition, many of the included studies did not adequately demonstrate the accuracy of the constructed MSP. Future prospective studies with an emphasis on the impact, accuracy, and clinical applicability of MSP construction techniques in cases of facial asymmetry are required.