Volumetric Measurement of Alveolar Clefts for Bone Graft Planning: A Systematic Review

2.1. Eligibility Criteria

Inclusion criteria were cross-sectional research journal articles that reported defect volumetric size utilizing imaging data from CT or CBCT, focused on the protocols of preoperative volumetric measurements for alveolar bone graft in cleft cases, and were full text and written in English.

The exclusion criteria included studies involving syndromic cleft, gingiva- or periosteoplasty, and bone graft for purposes other than filling the alveolar defect in the cleft.

Studies were also excluded if they had no description of the workflow for volume measurement, involved simulated defect and animal models, and were published as review articles, case reports, case series, expert opinions, and conference poster abstracts. Additionally, protocols described in volumetric measurements to evaluate or compare clinical interventions and assessments of bone graft outcomes or post-operative assessments were considered a minor focus and, therefore, excluded.

2.2. Search Strategy

Searches were conducted on PubMed, Embase, and Scopus databases in November 2023, with access provided through the author’s institution library access. Search strategies are listed in Table 1, which were optimised based on the rules for the search engine of each database to obtain the most relevant articles and the least amount of non-relevant articles. MesH terms in PubMed and Emtree in Embase were included in the search strategy. Additional relevant articles were also hand-searched through Google Scholar and reference lists. Titles and abstracts were screened by two examiners following inclusion and exclusion criteria. Afterward, full-text articles were retrieved and assessed for inclusion and exclusion by the same investigators. All reasons for exclusion were recorded. Consensus between the first, second, and third authors was sought if there was any ambiguity.

2.3. Data Extraction

The articles were noted for their inclusion and exclusion criteria, samples, patient characteristics, cleft type, study results, modality, workflow, anatomical landmarks of the defect borders, software and hardware, and reliability assessment by two investigators. The workflows were summarized by dividing them into several general types. They were all summarised, tabulated, noted if missing, presented, and discussed in the paper.

2.4. Study Quality Assessment

The quality appraisal tool, JBI Critical Appraisal Checklist for Analytical Cross-Sectional Studies [12], was used for quality assessment. Some of the parameters were modified for the study to clarify essential requirements for a robust and reproducible measurement protocol, specifically including assessments of the gold standard, reliability, and objective anatomical landmarks.

Q4 was considered positively assessed only if landmarks for the superior, inferior, and buccopalatal borders were clearly described. If either one or two sides of the borders were missing, they were considered unclear. Q7 received a positive assessment if all steps were specified with clarity that allowed for easy and unambiguous repetition. If there was no description of workflow, the study was excluded as part of the exclusion criteria. If multiple methods were addressed in the study, all methods were considered in unison for quality assessment.

3.1. Study Selection

A total of 1,594 titles and abstracts were imported into the Zotero reference manager (version 6.0.27) to be manually reviewed for duplicate removal. One study was flagged and retracted due to self-plagiarism, resulting in 233 articles. After reviewing titles and abstracts based on inclusion and exclusion criteria, 202 articles were removed. A total of 31 studies were read in full text, and 14 of them were excluded based on the same criteria. At the end of the study, 17 articles were included. The search process diagram is presented in Fig. (1).

3.2. Study Quality Assessment

The most common quality issues identified were a lack of comparison to the gold standard (only 24% of the studies were positively assessed in Q3a), lack of inter- and intraobserver reliability (29% in Q3b), and incomplete anatomical landmarks (24% in Q4) (Fig. 2). It is worth noting that three studies failed to report reliable results despite having multiple observers/measurements, leading to questions about selective reporting [13-15].

3.3. Study Characteristics

A summary of the sampling methods, patient characteristics, cleft types, and main findings from all included studies is presented in Table 2. The average defect volume for unilateral cleft only, bilateral cleft only, and all types of clefts were 1.38 cm³, 1.72 cm³, and 1.61 cm³, respectively. Most studies did not differentiate based on the involvement of the palate; therefore, the volumetric difference between CL and CLP could not be evaluated.

Table 3 summarises technical details, including modality, software and hardware, type of workflow, and anatomic landmarks. A total of 10 studies used CBCT [10, 15-23], and 7 studies used CT [5, 13, 14, 24-27]. Thirty workflows for alveolar defect volume measurement were found among 17 studies. Measurement by experts can be broadly divided into virtual and 3D printing methods. The most often used workflow for virtual measurement was manual segmentation in each axial slice [5, 13, 15, 16, 18, 20, 21, 23-27], followed by orthogonal tracing or editing [15, 17], mirrored template using automatic superimposition [14, 27], geometric estimation [22], region growing [19], and custom algorithms [19]. Most of the studies used commercial medical image segmentation, analysis, and design software (Mimics, Materialise, Leuven, Belgium) for virtual measurements. With 3D printing, 6 studies printed the patient’s maxilla [13, 14, 17, 21, 24, 26], while one study made a physical model of the defect itself [19]. One additional study investigated the use of artificial intelligence through a 3D U-Net architecture for segmentation and measurement [10].

Various anatomical landmarks were described in the literature (Table 3). Six studies provided a complete description of virtual landmarks for superior, inferior, and buccopalatal boundaries of the alveolar cleft [5, 16, 17, 21, 23, 27]. Six studies only included partial virtual landmarks [10, 15, 19, 20, 25, 26], whereas 5 others offered none [13, 14, 18, 22, 24]. Only one landmark description was described for 3D printing [21]. None were needed for AI.

Additionally, 4 studies showed comparisons to actual graft volume, which varied from measuring the actual graft intra-operatively [5, 15] to making impressions during surgery [18] or imaging 1 month after surgery [10]. Intraclass correlation coefficient (ICC) measurements for reliability ranged from 0.305 – 0.80.

4.1. Evaluation of Workflow for Virtual Measurements

The most commonly used workflow involved manually tracing the defect boundaries slice by slice in the axial view [5, 13, 16, 18, 20, 23-25, 27]. This technique is simple although possibly tedious, but can be achieved by a lot of software programs. Manual tracing has been tested against actual bone volume with clinically acceptable accuracy. Due to the manual nature of this technique, reproducibility is of particular concern. The studies utilising various software reported reliability ICC values ranging from 0.626-0.980 [5, 16, 20, 21, 23, 24, 26, 27].

An open-sourced software (Invesalius, CTI, Campinas, Brazil), which tends to be harder to use, was an outlier [16], presenting significantly worse reliability. Excluding this software, the reliability for axial tracing was at least 0.80. This indicates that the choice of software must also be considered, and it is best to use software programs that have been validated for clinical use.

Based on the findings in this review, commercial medical image processing software (Mimics, Materialise, Leuven, Belgium) offers multiple methods and has a reasonable amount of supporting evidence. Multiple-slice tracing was also used to accelerate the measurement process. However, both studies that used multiple-slice tracing did not mention the spacing distance between traced slices. Nevertheless, tracing in the axial slice seemed to be an adequately reliable method based on available evidence, although more studies are needed to confirm its accuracy.

Two studies evaluated a dental implant planning software (SimPlant Pro, Dentsply Sirona, Charlotte, USA) using manual orthogonal tracing or editing [15, 17]. Reported reliability was suboptimal, with intraclass correlation coefficients ranging from 0.650 to 0.838. However, the workflow and anatomical landmarks in the virtual method were described with good clarity. Based on these results and missing reliability data, orthogonal tracing or editing with this implant planning software could not be advocated at present due to the lack of evidence.

Mirroring the maxilla to use as a template in unilateral cases was explored in two studies, and both used automated superimposition [14, 27]. A personalised template made from the patient’s anatomy has the potential to reduce operator variability in shaping optimal alveolar shape in filling the defect, especially using automated superimposition. However, asymmetry of the nasal pyriform associated with unilateral clefts potentially complicates superimposition. Although there is no comparison to actual graft volume, this review found comparable accuracy to tracing slice-by-slice and 3D printing [14, 27]. However, reporting must be improved, especially in terms of reliability and comparison to actual graft volume.

Quereshy et al. measured defect width, height, and thickness to estimate volume using a geometric formula [22]. Intraobserver ICC was poor for defect thickness (0.305). Workflow and landmark anatomy were both unclear, which may have contributed to decreased reliability. Additionally, the measurements seemed to have been identified from a 3D-rendered virtual model rather than from sliced images, which decreases accuracy.

Barbosa et al. also found that simple geometric estimation is significantly inaccurate compared to true volumetric measurements [28]. Consequently, this technique should be avoided.

Semi-automatic techniques were used in two studies [10, 19]. Kasaven et al. explored a region-growing tool and a custom algorithm with industrial and materials science image analysis software (Volume Graphics Studio Max, Hexagon, Stockholm, Sweden) and a programmable image processing and analysis software (MATLAB, The MathWorks, Inc., Natick, USA), respectively [19]. These two techniques are highly dependent on density boundaries; thus, they have limited suitability to cleft images and require significant manual refinement. Wang et al. used a combination of manual and semi-automatic processes in an open-source medical image segmentation software (ITK-Snap, Penn Image Computing and Science Laboratory, Philadelphia, USA) [10]. However, the lack of clarity in the workflow described by Wang et al. left some uncertainty in the exact tools used. The study reported 10 hours of work per CBCT, which is a considerable difference from the minutes of work reported with other software (Mimics, Materialise, Leuven, Belgium) [16]. Therefore, time versus cost might be an important consideration in deciding which software to use. Currently, there is not enough evidence to promote these semi-automatic techniques for clinical purposes.

Wang et al. also investigated AI technology by using 3D U-Net, an AI architecture based on convolutional neural network [10]. The architecture was designed to work with small training datasets and large images, specifically to segment medical images [38]. During the training of the AI system, Wang et al. split each CBCT image into patches in the sliding window technique to be trained and create an output of labels for the segmentation. Optimization was done via generalised dice loss [10]. The similarity of AI segmentation compared to manual segmentation was 0.77 +/- 0.06, which was considered moderate to high [39]. Several other studies have also recently used AI for cleft [40, 41]. The use of AI for identifying and measuring the volume of alveolar cleft is promising, and further development is needed. It should be noted that to have any clinical value. AI systems must be developed using evidence-based data and applied under the supervision and approval of clinicians.

4.2. Evaluation of Workflow for 3D Printing

3D printing itself has been established to have adequate accuracy for medical and dental use [42, 43]. For alveolar cleft measurement, typically, plasticine was filled by observers into a 3D-printed skull. Thus, measurement was based on the plasticine filling rather than any 3D-printed object. The printed skull itself may be useful for education or precise, personalised planning [44, 45]. An exception to this was the research by Kasaven et al., in which a defect-shaped object was printed. This protocol could create 3D-printed biomaterial or scaffold [46] to aid alveolar bone graft surgery in addition to mere measurement. The time required for the 3D printing method was consistently longer than virtual measurements [13, 24, 26], which could be a significant drawback. Although cost was not part of the research from any of the included studies, it should also be part of consideration when choosing the technique.

In terms of cleft volume measurement by 3D-printed methods, this review found comparable accuracy to many types of virtual workflow, such as axial tracing [13, 21, 24, 26], orthogonal tracing [17], mirrored template [14], and region-growing techniques [19]. Yet, no studies were found that compared 3D printing-based cleft measurement to actual bone graft volume. Reported reliability ranged from 0.765 to 0.949 [19, 21, 24], although two studies did not provide reliable data [13, 14]. One study found moderate interobserver reliability between 2 cleft surgeons [17]. This was possibly due to the absence of defined landmarks for cleft filling in all 3D-printed protocols, except for the one proposed by Phienwej et al., who created superior and inferior boundaries for the defect, printed as part of the skull model [21]. While this attempt was appreciated, such physical boundaries would not be useful if the maxilla model is to be used as a simulation for surgery due to altered anatomy. Even though the research was done on 3D-printed objects, the fact that suboptimal reliability was found between cleft surgeons using physical models [17] could be significant. An argument could be made that this unreliability could extend to the surgery itself, supported by studies that found operator dependency of alveolar bone graft surgery [47, 48] and evidence of under- or overfilling [4, 5].

Additionally, literature on alveolar bone graft surgery frequently gives little description of landmarks or guidance for adequate defect filling. This highlights the importance of preoperative cleft volume measurements to guide surgeons and reduce operator dependency. Nonetheless, 3D printing for cleft care is promising and warrants future exploration with particular regard to landmark description and reliability.

4.3. Landmarks

Landmarks to determine borders of the alveolar defect should preferably be reliable, fast, and easy to determine. More importantly, they should have clinical significance and be compatible with good functional and aesthetic outcomes. Landmarks chosen for pre-operative determination of defect borders should reflect landmarks seen during surgery. However, none of the studies that provided a complete protocol for anatomical landmarks in this review had a comparison to actual bone volume.

In the current review, five studies used Cemento-enamel Junction (CEJ) [16, 17, 23, 26, 27], and another four used alveolar ridge [5, 15, 20, 21] as landmarks for the inferior margin. The normal alveolar ridge is 1-2 mm below CEJ and even farther in periodontitis or in cases of the alveolar cleft with bone dehiscence. CEJ has the advantage of being relatively constant and easily recognizable in imaging. Normal bone height and CEJ are both used as benchmarks for success evaluation [6, 49-52] and are agreeable with one another [53]. Therefore, CEJ may be preferable as a landmark for the inferior margin of alveolar defect.

For the superior margin, four studies used landmarks associated with the Anterior Nasal Spine (ANS) [15-17, 20], and two studies used a base of pyriform aperture [5, 26]. Phienwej et al. used both landmarks, describing the superior limit as a plane defined by ANS, the most lateral and inferior point of the pyriform aperture on the cleft side, and the most superior point of the mesiopalatal margin of the lateral alveolar segment [21]. Liu et al. set the superior limit at 3 mm below ANS [27]. Both ANS and pyriform aperture have been reported in the literature to be used as surgical landmarks and success measures [6, 54-58]. In cleft cases, the base of the pyriform aperture is harder to identify because it is absent in a complete alveolar defect and often asymmetrical compared to the normal side in unilateral clefts. ANS, as the superior limit, has the most supporting evidence and may be more desirable.

Alveolar thickness and shape may be of concern in the labio-palatal dimension. Three studies followed the contour of the contralateral side [10, 23, 27], two studies followed the continuity of the maxillary arch [16, 17], and another two studies used straight lines connecting the alveolar segments [21, 25]. One study prioritised thickness by dictating the defect filling to have equal thickness with adjacent alveolar segments [5]. However, not many studies evaluating volumetric ABG outcomes have described labio-palatal margins. Nagashima et al. used a mirror image of the contralateral side as a benchmark for the outcome, while Kibe et al. used a tangent line connecting the labial and palatal edges of the defect [31, 59]. As long as good continuity of the arch is achieved, literature is sparse to determine if graft shape (contoured versus straight line) affects functional or facial aesthetic outcomes. Additionally, the choice between ideal contour and straight lines might also be practical as some software may not be able to create contoured lines.

Furthermore, the majority of the studies have used graft thickness as a success measure. Padwa et al. determined graft success if the labio-palatal thickness of the graft is ≥75% of adjacent tooth root width [6]. Stasiak et al. determined a moderate outcome if the thickness is ≥50% - ≤100% of central incisor root width and a good outcome if the thickness is ≥100% [30]. Ideally, it should be at least 8 mm thick to accommodate the size of canines [29]. To comply with the outcome assessment, we recommend prioritizing good continuity with the arch and appropriate thickness as labiopalatal guidance.

Creating planes for defect limit would be more time-consuming than point landmarks, and current evidence does not clearly indicate whether it improves reliability. Point landmarks would be faster and easier to create but are more prone to patient positioning errors. One study reported that non-oriented and oriented data had equal performance [23]. With current studies, there is no evidence to discourage the use of point landmarks.

Among articles that used 3D printing, only one study attempted to provide some guidance for identifying the borders of the filling [21]. Planning the operational definition of anatomical landmarks for the defect borders is important to obtain the most reliable measurements. Although most of the reported reliability was good to excellent, a lot of the inter- and/or intraobserver reliability data in these studies were incomplete. In this regard, standardisation of border landmarks would create comparability between studies. Therefore, future studies with more thorough reporting of the landmarks and reliability are needed.

4.4. Limitations

Several studies were identified that included indirect comparisons to actual bone volume, but the validity of these appointed gold standard measurements was out of the scope of this study. Factors such as modality, X-ray parameters, voxel size, and slice thickness were also not evaluated in detail. However, evidence has shown that the choice of modality (CT and CBCT) and parameters do not significantly affect measurements of large structures, such as cleft [34, 60]. While some aspects of treatment flow (such as time and cost) were briefly discussed in the review, this aspect is rarely discussed in the studies, especially in regard to the surgery itself. Moreover, further research is needed to address important deficiencies in the current literature, such as investigating the validity and accuracy of each measurement method, reliability, and standardising landmarks to ensure volume comparability between studies.

Furthermore, the influence of preoperative volume determination on treatment flow and surgical consistency must be covered in areas of investigation. Recommendations in this review are based on available literature and professional judgement from experts with decades of experience; however, some aspects still require further evidence. Other factors that may impact surgical success and consistency are beyond the scope of this review. Nevertheless, the findings of this review would be important to pinpoint gaps in current knowledge and establish more standardized landmarks and protocols for future studies.