This article summarizes the current state of diagnostic modalities for infant craniofacial deformities and highlights capable diagnostic tools available currently to pediatricians.

Video Abstract

Video Abstract

Close modal
CONTEXT

Deformational plagiocephaly is a common diagnosis encountered by pediatricians in the first year of life. Subjective clinical examination and documentation is the most common method for identifying and monitoring the evolution of head shape and determining treatment or the decision to refer to a specialist.

OBJECTIVE

In this systematic review (PROSPERO; CRD42021224842), we aim to compile the evidence for non-radiographic screening and monitoring modalities for deformational plagiocephaly in infants to support pediatricians in achieving earlier diagnosis and more objective monitoring.

DATA SOURCES

A systematic review of the literature was conducted using OVID Medline, Embase, and Web of Science.

STUDY SELECTION

Articles pertaining to the use or evaluation of diagnostic modalities for plagiocephaly in infants published between January 1990 and August 2021.

DATA EXTRACTION

Data on diagnostic accuracy, time-to-diagnosis, reliability, and outcomes for each modality were collected as available by two independent reviewers.

RESULTS

A total of 22 studies were included. We identified 5 unique head shape monitoring technologies: anthropometry, plagiocephalometry, 3D laser scanning, digital photographic, and 3D photogrammetry. Smartphone and artificial intelligence integration have increased in plagiocephaly and craniosynostosis screening and monitoring tools.

LIMITATIONS

Inconsistent reporting both inter- and intra-modality hindered meaningful comparison between screening tools. Substantial heterogeneity in measured outcomes, study design, and population size made cross-study comparisons difficult.

CONCLUSIONS

A growing list of quantitative diagnostic modalities for head shape monitoring are becoming accessible to pediatricians. The introduction of artificial intelligence to 3D photogrammetry and digital photography with easy-to-use smartphone applications seems promising for future diagnostic efficiency.

Deformational plagiocephaly is the leading cause of head shape abnormalities in children, with mild cases affecting ∼40% of infants aged <1 year.1  Severe cases of deformational plagiocephaly are much less frequent, and in exceptional cases may require surgical intervention. Although it spares a child’s neurocognitive development, deformational plagiocephaly can cause significant cosmetic deformities, such as facial asymmetry, eventual malocclusion needing orthodontic treatment, and social stigmatization later in life.2  Most often, it is detected by the child’s parents or pediatrician through simple visual assessment.3  Subsequent monitoring of abnormal head shape is crucial because the progression of deformity, despite conservative treatment of deformational plagiocephaly, can be indicative of the need for additional intervention (such as treatment of persistent torticollis) or an undiagnosed sutural fusion (craniosynostosis) requiring surgical treatment.

The most common monitoring strategy remains clinical examination including documentation of head circumference. Combined with the interval between visits, the subjective nature of clinical examination for the monitoring of head shape improvement or deterioration remains a challenge. As such, determining if and when to refer patients for additional treatment (physiotherapy for persistent torticollis) or specialty consultation (persistent plagiocephaly, possible craniosynostosis) remains a therapeutic dilemma.

To that end, there are a number of new, readily available imaging modalities that may provide pediatricians with objective measures of head shape progression and assist in diagnostic decision-making. Technological advances in the medical field, bolstered by parallel advancements in hardware, computing power, and artificial intelligence (AI) capability, have introduced new diagnostic methods capable of tracking head shapes.4  Older quantitative methods of monitoring head asymmetry demonstrated acceptable accuracy but were expensive and/or time-consuming because calculations and postprocessing were primarily manual. To our knowledge, no recent comprehensive review of the screening and monitoring modalities for deformational plagiocephaly exists to highlight progress in time-to-diagnosis, accuracy, and accessibility. The aim of this study is to compile and evaluate the full complement of diagnostic modalities available to support pediatricians in identifying and monitoring cranial deformity in infants. The overarching goal involves eventual implementation of capable diagnostic support tools to ultimately help guide appropriate treatment and specialist referral.

This systematic review adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines and was prospectively registered in PROSPERO (CRD42021224842).

The review included studies involving children or models of children aged <18 years, with a primary focus on infants aged <12 months. Studies involved comparison or evaluation of diagnostic techniques or technologies (nonradiographic modalities), new diagnostic software analysis systems, pilot/proof-of-concept studies for new diagnostic modalities, or studies comparing diagnostic measurements to determine the most useful head shape parameters. There was no restriction to the type of study design, and only English language studies were considered.

Studies were excluded if they met any of the following criteria: case-report, focus on a diagnostic technique involving radiation, ultrasound, focus on surgical recovery, surgical outcome evaluation, genetic-based diagnoses, prenatal diagnosis, and intracranial volume quantification. Conference proceedings and abstracts were also excluded.

A literature search was conducted on November 11, 2020, in Ovid Medline, Ovid Embase, and Web of Science. The search strategy was built using subject headings, keywords, and Medical Subject Headings terms related to “diagnosis,” “screening,” “monitoring,” “plagiocephaly,” and “craniosynostosis.” These guiding terms were chosen as those most likely to recommend clinically useful tools and developments for pediatricians. The full search strategy for Ovid Medline and Ovid Embase is detailed in Supplemental Information. The search encompassed all publications between January 1990 and August 2021 because developments explored before 1990 would no longer be relevant given the pace of technological advancement.

Records obtained from the initial search strategy (13 857 records) were imported into Endnote ×9 for deduplication. The de-duplicated results were imported into Rayyan for screening. Two independent reviewers (A.W. and D.Z.) screened each study for relevance based on title and abstract (level 1 screening). Conflicting decisions were resolved by discussion and consensus among the authors. The full text of each study included after level 1 screening was then acquired and reviewed by the authors to determine final inclusion or exclusion. Any conflicts were resolved by discussion and consensus of the authors.

We followed a descriptive synthesis approach and categorized the included studies by diagnostic modality. Data on diagnostic accuracy, time-to-diagnosis, reliability, and technological novelty were collected independently by A.W. and D.Z. as available for each included study. Data were then summarized through 4 key diagnostic domains that best represented developments in the field: shape analysis, reliability, smartphone integration, and automation.

The systematic review identified 13 857 articles, of which 9349 articles remained after de-duplication and 62 articles remained after title and abstract screening. Full-text review returned a final inclusion of 22 articles spanning 5 diagnostic modalities (Fig 1) (Tables 1 and 2). The most frequent modality described in the included studies was photogrammetry (50%). Our review included studies describing the following modalities: photogrammetry (11), anthropometry (2), laser scanning (1), two-dimensional (2D) digital photography (7), and plagiocephalometry (PCM) (1) (Fig 2, Tables 1 and 2).

FIGURE 1

Summary of systematic search results

FIGURE 1

Summary of systematic search results

Close modal
FIGURE 2

A, Anthropometric measurement of the head using calipers. Plagiocephalometric measurement of the head using (B) a thermoplastic band and (C) sketching. D, Digital photographic measurement of head shape. E, Example of cap required for 3D analysis of cranial shape used in (F) 3D laser scanning, and (G) 3D photogrammetry. H, Example of cranial measurements obtained from 3D renders of an infant’s skull in multiple planes. Panels A through H are being reprinted with permission. (A, Reprinted with permission from Wilbrand JF, Wilbrand M, Pons-Kuehnemann J, et al. Value and reliability of anthropometric measurements of cranial deformity in early childhood. J Craniomaxillofac Surg. 2011;39(1):24–29. With permission from Elsevier. B, and C. van Adrichem L, van Vlimmeren L, Cadanová D, et al. Validation of a simple method for measuring cranial deformities (plagiocephalometry). J Craniofac Surg. 2008;19(1):15–21. With permission from Wolters Kluwer. D, Schaaf H, Wilbrand JF, Boedeker RH, et al. Accuracy of photographic assessment compared with standard anthropometric measurements in nonsynostotic cranial deformities. Cleft Palate Craniofac J. 2010;47(5):447–453. With permission from SAGE Publications, Inc. E, Barbero-García I, Lerma JL, Mora-Navarro G. Fully automatic smartphone-based photogrammetric 3D modeling of infant’s heads for cranial deformation analysis. J Photogramm Remote Sens. 2020;166:268–277. With permission from Elsevier. F, Nahles S, Klein M, Yacoub A, Neyer J. Evaluation of positional plagiocephaly: conventional anthropometric measurement versus laser scanning method. J Craniomaxillofac Surg. 2018;46(1):11–21. With permission from Elsevier. G, Schaaf H, Malik CY, Howaldt HP, Streckbein P. Evolution of photography in maxillofacial surgery: from analog to 3D photography - an overview. Clin Cosmet Investig Dent. 2009;1:39–45. Open access copyright. H, Aarnivala H, Vuollo V, Heikkinen T, et al. Accuracy of measurements used to quantify cranial asymmetry in deformational plagiocephaly. J Craniomaxillofac Surg. 2017;45(8):1349–1356. With permission from Elsevier.)

FIGURE 2

A, Anthropometric measurement of the head using calipers. Plagiocephalometric measurement of the head using (B) a thermoplastic band and (C) sketching. D, Digital photographic measurement of head shape. E, Example of cap required for 3D analysis of cranial shape used in (F) 3D laser scanning, and (G) 3D photogrammetry. H, Example of cranial measurements obtained from 3D renders of an infant’s skull in multiple planes. Panels A through H are being reprinted with permission. (A, Reprinted with permission from Wilbrand JF, Wilbrand M, Pons-Kuehnemann J, et al. Value and reliability of anthropometric measurements of cranial deformity in early childhood. J Craniomaxillofac Surg. 2011;39(1):24–29. With permission from Elsevier. B, and C. van Adrichem L, van Vlimmeren L, Cadanová D, et al. Validation of a simple method for measuring cranial deformities (plagiocephalometry). J Craniofac Surg. 2008;19(1):15–21. With permission from Wolters Kluwer. D, Schaaf H, Wilbrand JF, Boedeker RH, et al. Accuracy of photographic assessment compared with standard anthropometric measurements in nonsynostotic cranial deformities. Cleft Palate Craniofac J. 2010;47(5):447–453. With permission from SAGE Publications, Inc. E, Barbero-García I, Lerma JL, Mora-Navarro G. Fully automatic smartphone-based photogrammetric 3D modeling of infant’s heads for cranial deformation analysis. J Photogramm Remote Sens. 2020;166:268–277. With permission from Elsevier. F, Nahles S, Klein M, Yacoub A, Neyer J. Evaluation of positional plagiocephaly: conventional anthropometric measurement versus laser scanning method. J Craniomaxillofac Surg. 2018;46(1):11–21. With permission from Elsevier. G, Schaaf H, Malik CY, Howaldt HP, Streckbein P. Evolution of photography in maxillofacial surgery: from analog to 3D photography - an overview. Clin Cosmet Investig Dent. 2009;1:39–45. Open access copyright. H, Aarnivala H, Vuollo V, Heikkinen T, et al. Accuracy of measurements used to quantify cranial asymmetry in deformational plagiocephaly. J Craniomaxillofac Surg. 2017;45(8):1349–1356. With permission from Elsevier.)

Close modal
TABLE 1

Summary of Included Studies

ModalityAuthor, YearStudy DesignSample SizeReference StandardFindings
Anthropometry, standardized protocol Wilbrand et al, 201125  Diagnostic validation study n = 30 Expert measurement Mean intraobserver variability <1.131 mm2, interobserver variability <0.182 mm2. Overall 2 mm measurement variance. 
Anthropometry Wu et al, 202012  Retrospective diagnostic validation study n = 89 CT No statistical difference between caliper measurements and CT (P > .05). Anterior–posterior caliper dimensions were within 1 cm of CT in 73% of cases, whereas 88% of transverse measurements were within 1 cm of CT. 
Digital photography/machine learning Callejas Pastor et al, 202019  Retrospective diagnostic validation study n = 80 3D-cranial scanning 86.7% classification accuracy for brachycephaly and plagiocephaly. Cephalic ratio and CVAI correlation coefficients were 0.85 and 0.89, respectively. 
Digital photography Schaaf H. et al, 201026  Diagnostic validation study n = 122 Anthropometry CI CVAI intraclass correlation coefficients were 0.982 and 0.946, respectively. Digital photography satisfied the limits of agreement (CI, 7.51%; CVAI 6.57%). 
Digital photography Lopes Alho et al, 202016  Diagnostic validation study (case control) n = 2 None SymMetric can differentiate between control and plagiocephalic patients in the superior view and detected clinical improvement after orthotic use. 
Digital photography/ machine learning Agarwal et al, 201820  Diagnostic validation study (augmented database) n = 1006 images (75:25 training: test distribution) Professionally classified images AUC of 0.95 for cleft abnormality and craniosynostosis. Validation accuracy of model is 92.22%, testing accuracy is 84.12%. 
Digital photography/ machine learning Bookland et al, 202127  Retrospective diagnostic validation study n = 339 retrospectively collected cranial images, 40 open-source cranial images Professionally classified images/ optical scan-derived craniometric measurements Cranial shape classification had an accuracy of 93.3% (95% CI 86.8–98.8; P < .001), with a sensitivity of 92% and a specificity of 94.3%. Intraclass correlation coefficients for measurements of the CI and CVAI compared with optical measurements were 0.95 (95% CI 0.84–0.98; P < .001) and 0.67 (95% CI 0.24–0.88; P = .003). 
Digital photography/ machine learning Geisler et al, 202128  Diagnostic validation study n = 1076 images Professionally classified images ResNet-50 CNN obtained an overall accuracy of 90.6% for diagnosing craniosynostosis. Sensitivity and precision for a combined top-front view were 100% and 100% for metopic synostosis, 93.3% and 100% for sagittal synostosis, and 66.7% and 100% for unicoronal synostosis. 
Digital photography Hutchison et al, 200521  Diagnostic validation study (case control) n = 31 Flexicurve ruler Oblique cranial length ratio >106% can define plagiocephaly, and CI > 93% can define brachycephaly. Photographic method was better tolerated by infants, more repeatable, and preferred by mothers. 
PCM van Adrichem et al, 200813  Diagnostic validation study n = 21 3D CT No statistical difference (P > .05) between PCM ring (removed from the head) and CT-based skull measurements. 
3D laser scanning Nahles et al, 201815  Diagnostic validation study n = 44 Anthropometry Mean head circumference was 441.5 mm for the anthropometric measurements and 441.6 mm for the laser scan method, with no significant difference between the 2 methods. A significant difference was found regarding the head width (P < .001), head length (P < .05), diagonals (P < .001), and distance ex-t (P < .001). Mean scan time for scanning was 579.6 s in contrast to 180.5 s for the manual anthropometric method. 
3D photography Aarnivala et al, 20176  Diagnostic validation study n = 407 images Expert rating with Argenta classification Oblique cranial length ratio consistently provided the best discrimination in terms of 3D imaging area under the curve values. Optimal cutoff values for deformational plagiocephaly (Argenta class ≥ 1) across all age groups were 104.0% for oblique cranial length ratio (83% sensitivity, 97% specificity), 10.5% for posterior cranial asymmetry index (90% sensitivity, 90% specificity), and 24.5 for weighted asymmetry score (88% sensitivity, 90% specificity). 
3D photography Schaaf et al, 20107  Retrospective diagnostic validation study n = 100 Anthropometry Comparison of the 3D photographic and calipers measurements showed that 3D photography resulted in a slight overestimation. Interrater reliability was 0.97 for plagiocephaly and 0.98 for brachycephaly. 
3D photography Skolnick et al, 20148  Retrospective diagnostic validation study n = 26 Nonlinear measures from 3D photographs The linear measure that best correlated with the inclusive measures of asymmetry was FZ-EU, the distance from the frontozygomaticus to the contralateral eurion (r ≥ 0.90). Correlations between measures (0.10<r < 0.95) and intrarater reliability (correlation coefficients from 0.42 to 0.99) of linear measurements varied widely. 
3D photography/machine learning Porras et al, 201922  Retrospective diagnostic validation study n = train: 201, test: 18 CT The algorithm detected craniosynostosis automatically with 94.74% sensitivity and 96.02% specificity. It further correctly identified the fused sutures with 99.51% sensitivity and 99.13% specificity. 
3D photography/machine learning de Jong et al, 202023  Retrospective diagnostic validation study n = 213 CT 195 out of 196 3D stereophotographs (99.5%) were correctly classified for a craniosynostosis diagnosis by the deep learning algorithm. 
3D photography Barbero-García et al, 202011  Retrospective diagnostic validation study n = 5 CT/MRI CT/MRI confirmed accuracy below 1.5 mm. Basic, automatically derived, anthropometric-linear magnitudes obtained a mean variability of 0.6±0.6 mm for the longitudinal and transversal distances and 1.4±1.3 mm for the maximum perimeter. 
3D photography Barbero-García et al, 201818  Diagnostic validation study n = 10 CT/MRI Smartphone-based 3D photogrammetric models overestimated measurements by up to 3.2 mm because of both hair and usage of caps. Differences in shape are below 1.5 mm for every patient. 
3D photography/machine learning Barbero-García et al, 202114  Diagnostic validation study n = 5 Target-based coded markers on a cap Precision of cap points in the generated 3D point cloud is close to 1 mm. Eye detection returned an SD around 2 mm; mouth and nose detection had SDs of 8.1 and 5.7 mm, respectively. 
3D photography Meulstee et al, 20179  Diagnostic validation study n = 100 CT Principal component analysis was used to find the mean cranial shape and the cranial shape variation in the normal population. The model distinguished scaphocephaly (P < .001) and trigonocephaly (P > .001) patients from the normal population. 
3D photography/machine learning Tu et al, 201924  Diagnostic validation study n = 28 CT The trained support vector machine classifier obtained an improved accuracy of 91.03% in the detection of craniosynostosis, compared with 78.21% obtained using head circumference or CI. 
3D photography Atmosukarto et al, 201010  Diagnostic validation study n = 254 Expert rating Novel 3D-based plagiocephaly posterior severity scores provided better sensitivity and specificity in the discrimination of plagiocephalic and typical head shapes than the 2D measurements provided by a close approximation of oblique cranial length ratio. AUC statistics were as follows: left posterior flattening score (97%), right posterior flattening score (91%), asymmetry score (99%), absolute asymmetry score (91%), and approximation of a previously described 2D measure, the oblique cranial length ratio (79%). 
ModalityAuthor, YearStudy DesignSample SizeReference StandardFindings
Anthropometry, standardized protocol Wilbrand et al, 201125  Diagnostic validation study n = 30 Expert measurement Mean intraobserver variability <1.131 mm2, interobserver variability <0.182 mm2. Overall 2 mm measurement variance. 
Anthropometry Wu et al, 202012  Retrospective diagnostic validation study n = 89 CT No statistical difference between caliper measurements and CT (P > .05). Anterior–posterior caliper dimensions were within 1 cm of CT in 73% of cases, whereas 88% of transverse measurements were within 1 cm of CT. 
Digital photography/machine learning Callejas Pastor et al, 202019  Retrospective diagnostic validation study n = 80 3D-cranial scanning 86.7% classification accuracy for brachycephaly and plagiocephaly. Cephalic ratio and CVAI correlation coefficients were 0.85 and 0.89, respectively. 
Digital photography Schaaf H. et al, 201026  Diagnostic validation study n = 122 Anthropometry CI CVAI intraclass correlation coefficients were 0.982 and 0.946, respectively. Digital photography satisfied the limits of agreement (CI, 7.51%; CVAI 6.57%). 
Digital photography Lopes Alho et al, 202016  Diagnostic validation study (case control) n = 2 None SymMetric can differentiate between control and plagiocephalic patients in the superior view and detected clinical improvement after orthotic use. 
Digital photography/ machine learning Agarwal et al, 201820  Diagnostic validation study (augmented database) n = 1006 images (75:25 training: test distribution) Professionally classified images AUC of 0.95 for cleft abnormality and craniosynostosis. Validation accuracy of model is 92.22%, testing accuracy is 84.12%. 
Digital photography/ machine learning Bookland et al, 202127  Retrospective diagnostic validation study n = 339 retrospectively collected cranial images, 40 open-source cranial images Professionally classified images/ optical scan-derived craniometric measurements Cranial shape classification had an accuracy of 93.3% (95% CI 86.8–98.8; P < .001), with a sensitivity of 92% and a specificity of 94.3%. Intraclass correlation coefficients for measurements of the CI and CVAI compared with optical measurements were 0.95 (95% CI 0.84–0.98; P < .001) and 0.67 (95% CI 0.24–0.88; P = .003). 
Digital photography/ machine learning Geisler et al, 202128  Diagnostic validation study n = 1076 images Professionally classified images ResNet-50 CNN obtained an overall accuracy of 90.6% for diagnosing craniosynostosis. Sensitivity and precision for a combined top-front view were 100% and 100% for metopic synostosis, 93.3% and 100% for sagittal synostosis, and 66.7% and 100% for unicoronal synostosis. 
Digital photography Hutchison et al, 200521  Diagnostic validation study (case control) n = 31 Flexicurve ruler Oblique cranial length ratio >106% can define plagiocephaly, and CI > 93% can define brachycephaly. Photographic method was better tolerated by infants, more repeatable, and preferred by mothers. 
PCM van Adrichem et al, 200813  Diagnostic validation study n = 21 3D CT No statistical difference (P > .05) between PCM ring (removed from the head) and CT-based skull measurements. 
3D laser scanning Nahles et al, 201815  Diagnostic validation study n = 44 Anthropometry Mean head circumference was 441.5 mm for the anthropometric measurements and 441.6 mm for the laser scan method, with no significant difference between the 2 methods. A significant difference was found regarding the head width (P < .001), head length (P < .05), diagonals (P < .001), and distance ex-t (P < .001). Mean scan time for scanning was 579.6 s in contrast to 180.5 s for the manual anthropometric method. 
3D photography Aarnivala et al, 20176  Diagnostic validation study n = 407 images Expert rating with Argenta classification Oblique cranial length ratio consistently provided the best discrimination in terms of 3D imaging area under the curve values. Optimal cutoff values for deformational plagiocephaly (Argenta class ≥ 1) across all age groups were 104.0% for oblique cranial length ratio (83% sensitivity, 97% specificity), 10.5% for posterior cranial asymmetry index (90% sensitivity, 90% specificity), and 24.5 for weighted asymmetry score (88% sensitivity, 90% specificity). 
3D photography Schaaf et al, 20107  Retrospective diagnostic validation study n = 100 Anthropometry Comparison of the 3D photographic and calipers measurements showed that 3D photography resulted in a slight overestimation. Interrater reliability was 0.97 for plagiocephaly and 0.98 for brachycephaly. 
3D photography Skolnick et al, 20148  Retrospective diagnostic validation study n = 26 Nonlinear measures from 3D photographs The linear measure that best correlated with the inclusive measures of asymmetry was FZ-EU, the distance from the frontozygomaticus to the contralateral eurion (r ≥ 0.90). Correlations between measures (0.10<r < 0.95) and intrarater reliability (correlation coefficients from 0.42 to 0.99) of linear measurements varied widely. 
3D photography/machine learning Porras et al, 201922  Retrospective diagnostic validation study n = train: 201, test: 18 CT The algorithm detected craniosynostosis automatically with 94.74% sensitivity and 96.02% specificity. It further correctly identified the fused sutures with 99.51% sensitivity and 99.13% specificity. 
3D photography/machine learning de Jong et al, 202023  Retrospective diagnostic validation study n = 213 CT 195 out of 196 3D stereophotographs (99.5%) were correctly classified for a craniosynostosis diagnosis by the deep learning algorithm. 
3D photography Barbero-García et al, 202011  Retrospective diagnostic validation study n = 5 CT/MRI CT/MRI confirmed accuracy below 1.5 mm. Basic, automatically derived, anthropometric-linear magnitudes obtained a mean variability of 0.6±0.6 mm for the longitudinal and transversal distances and 1.4±1.3 mm for the maximum perimeter. 
3D photography Barbero-García et al, 201818  Diagnostic validation study n = 10 CT/MRI Smartphone-based 3D photogrammetric models overestimated measurements by up to 3.2 mm because of both hair and usage of caps. Differences in shape are below 1.5 mm for every patient. 
3D photography/machine learning Barbero-García et al, 202114  Diagnostic validation study n = 5 Target-based coded markers on a cap Precision of cap points in the generated 3D point cloud is close to 1 mm. Eye detection returned an SD around 2 mm; mouth and nose detection had SDs of 8.1 and 5.7 mm, respectively. 
3D photography Meulstee et al, 20179  Diagnostic validation study n = 100 CT Principal component analysis was used to find the mean cranial shape and the cranial shape variation in the normal population. The model distinguished scaphocephaly (P < .001) and trigonocephaly (P > .001) patients from the normal population. 
3D photography/machine learning Tu et al, 201924  Diagnostic validation study n = 28 CT The trained support vector machine classifier obtained an improved accuracy of 91.03% in the detection of craniosynostosis, compared with 78.21% obtained using head circumference or CI. 
3D photography Atmosukarto et al, 201010  Diagnostic validation study n = 254 Expert rating Novel 3D-based plagiocephaly posterior severity scores provided better sensitivity and specificity in the discrimination of plagiocephalic and typical head shapes than the 2D measurements provided by a close approximation of oblique cranial length ratio. AUC statistics were as follows: left posterior flattening score (97%), right posterior flattening score (91%), asymmetry score (99%), absolute asymmetry score (91%), and approximation of a previously described 2D measure, the oblique cranial length ratio (79%). 

AUC, area under the curve; CNN, convolutional neural network; ex-t, orbito-tragial distance; FZ-EU, frontozygomaticus to eurion; 95% CI, 95% confidence interval.

TABLE 2

Summary of Included Modalities

ModalitySummaryAdvantagesDisadvantagesComparable to Gold Standard (3D-CT)Clinical Availability
Anthropometry Anthropometry uses calipers and measuring tapes to take assorted measurements of the child’s head. These measurements are then used to calculate cranial asymmetry indices (CVAI, CI, OCLR, etc). • Simple
• Rapid
• Quantitative
• Low cost
• <2 mm inter/intrarater variability with standardized protocol 
• Requires calipers
• Subjective landmark identification
• 2D (cannot measure diagonal and vertical cranial dimensions)
• Requires standard protocol for good reliability. 
Yes, no statistically significant differences Currently available 
PCM PCM uses a thermoplastic band wrapped around the child’s head to make a mold that can be traced onto a paper. This trace can then be used to take measurements and calculate asymmetry indices without the child moving around. • Acceptable inter/intrarater reliability
• Low cost
• Mean difference between CT and PCM <1 mm. 
• Requires thermoplastic band and tracing
• Subjective landmark identification
• 2D (cannot measure diagonal and vertical cranial dimensions)
• More time-consuming than anthropometry 
Yes, no statistically significant differences Currently available 
Digital photography Most commonly, a digital photograph will be taken from the birds-eye-view perspective of the child’s head. Measurements can then be taken directly off the photo with manual landmark selection. Automated methods also exist that take a photo as input and return a diagnosis using automatically calculated asymmetry indices. • Rapid
• Reproducible
• Low variation
• Can use smartphone to take photo
• Can leverage AI for automated measurement and analysis of photos
• Photos can be reviewed at a later date for comparison.
• Not affected by patient age 
• Manual landmark selection is subjective.
• Automatic methods need larger training data sets to increase accuracy.
• 2D (cannot measure diagonal and vertical cranial dimensions)
• AI-based systems are limited by training data availability. 
N/A Yes, but portable smartphone-based tools require further refinement. 
3D photogrammetry 3D photogrammetry uses multicamera stationary imaging setups or, more recently, the slow-motion features on modern smartphones. A 3D image is pieced together from a large number of still photos taken from different perspectives. Multicamera setups typically require the child be fitted with a nylon cap. Cranial asymmetry can be evaluated using 2D anthropometric indices or more complicated 3D indices. • 3D representation of the head
• Rapid
• Smartphone-based methods are cheap
• Can leverage AI for automated measurement and analysis of head shape
• Can correctly identify the fused suture in craniosynostosis
• Multicamera installations are immune to infant movement (<1 s acquisition time).
• Not affected by patient age 
• Typically requires a nylon cap
• Multicamera installations are expensive and require a dedicated room and expert operation.
• Some smartphone-based methods can be susceptible to patient movement.
• AI-based systems are limited by training data availability. 
Yes, no statistically significant differences Yes, but limited to specialized centers (not accessible to pediatricians) 
3D laser scanning 3D laser scanning uses a stationary or handheld scanner to create a 3D representation of the infant’s head. The infant must be fitted with a nylon cap with reflective dots. 2D anthropometric or more complex 3D indices can be used to determine cranial asymmetry. • 3D representation of the head
• No advantages compared with anthropometric measurement when using 2D indices 
• Requires dot fixation with a nylon cap for 3D scanning.
• Significantly longer than anthropometric methods
• High acquisition, service, and maintenance cost
• Inconsistent results 
N/A Yes, but limited to specialized centers (not accessible to pediatricians) 
ModalitySummaryAdvantagesDisadvantagesComparable to Gold Standard (3D-CT)Clinical Availability
Anthropometry Anthropometry uses calipers and measuring tapes to take assorted measurements of the child’s head. These measurements are then used to calculate cranial asymmetry indices (CVAI, CI, OCLR, etc). • Simple
• Rapid
• Quantitative
• Low cost
• <2 mm inter/intrarater variability with standardized protocol 
• Requires calipers
• Subjective landmark identification
• 2D (cannot measure diagonal and vertical cranial dimensions)
• Requires standard protocol for good reliability. 
Yes, no statistically significant differences Currently available 
PCM PCM uses a thermoplastic band wrapped around the child’s head to make a mold that can be traced onto a paper. This trace can then be used to take measurements and calculate asymmetry indices without the child moving around. • Acceptable inter/intrarater reliability
• Low cost
• Mean difference between CT and PCM <1 mm. 
• Requires thermoplastic band and tracing
• Subjective landmark identification
• 2D (cannot measure diagonal and vertical cranial dimensions)
• More time-consuming than anthropometry 
Yes, no statistically significant differences Currently available 
Digital photography Most commonly, a digital photograph will be taken from the birds-eye-view perspective of the child’s head. Measurements can then be taken directly off the photo with manual landmark selection. Automated methods also exist that take a photo as input and return a diagnosis using automatically calculated asymmetry indices. • Rapid
• Reproducible
• Low variation
• Can use smartphone to take photo
• Can leverage AI for automated measurement and analysis of photos
• Photos can be reviewed at a later date for comparison.
• Not affected by patient age 
• Manual landmark selection is subjective.
• Automatic methods need larger training data sets to increase accuracy.
• 2D (cannot measure diagonal and vertical cranial dimensions)
• AI-based systems are limited by training data availability. 
N/A Yes, but portable smartphone-based tools require further refinement. 
3D photogrammetry 3D photogrammetry uses multicamera stationary imaging setups or, more recently, the slow-motion features on modern smartphones. A 3D image is pieced together from a large number of still photos taken from different perspectives. Multicamera setups typically require the child be fitted with a nylon cap. Cranial asymmetry can be evaluated using 2D anthropometric indices or more complicated 3D indices. • 3D representation of the head
• Rapid
• Smartphone-based methods are cheap
• Can leverage AI for automated measurement and analysis of head shape
• Can correctly identify the fused suture in craniosynostosis
• Multicamera installations are immune to infant movement (<1 s acquisition time).
• Not affected by patient age 
• Typically requires a nylon cap
• Multicamera installations are expensive and require a dedicated room and expert operation.
• Some smartphone-based methods can be susceptible to patient movement.
• AI-based systems are limited by training data availability. 
Yes, no statistically significant differences Yes, but limited to specialized centers (not accessible to pediatricians) 
3D laser scanning 3D laser scanning uses a stationary or handheld scanner to create a 3D representation of the infant’s head. The infant must be fitted with a nylon cap with reflective dots. 2D anthropometric or more complex 3D indices can be used to determine cranial asymmetry. • 3D representation of the head
• No advantages compared with anthropometric measurement when using 2D indices 
• Requires dot fixation with a nylon cap for 3D scanning.
• Significantly longer than anthropometric methods
• High acquisition, service, and maintenance cost
• Inconsistent results 
N/A Yes, but limited to specialized centers (not accessible to pediatricians) 

OCLR, oblique cranial length ratio

Two authors (A.W. and D.Z.) independently evaluated the risk of bias for all included studies using the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool (Supplemental Information).5  Of the 22 included studies, 19 were found to be at low risk of bias, whereas 3 were evaluated as high risk (Table 3).

TABLE 3

Risk of Bias Assessment Using QUADAS-2

Author, YearDomain 1Domain 2Domain 3Domain 4Overall
Q1Q2Q3ResultQ1Q2ResultsReference StandardQ2Q3ResultQ1Q2Q3Q4Result
Wilbrand et al, 201125  Low Low Expert measurement Low Low Low 
Wu et al, 202012  Low Low CT Low Low Low 
Callejas Pastor et al, 202019  Unclear Low 3D cranial scanning Low Low Low 
Schaaf H et al, 201026  Low Low Anthropometry Low Low Low 
Lopes Alho et al, 202016  High High None High High High 
Agarwal et al, 201820  Low Low Professionally classified images Low Lowa Low 
Bookland et al, 202127  Low Low Professionally classified images/optical scan-derived craniometric measurements Low Low Low 
Geisler et al, 202128  Low Low Professionally classified images Low Low Low 
Hutchison et al, 200521  Low Low Flexicurve ruler Low Low Low 
van Adrichem et al, 200813  Low Low 3D CT Low Low Low 
Nahles et al, 201815  Low Low Anthropometry Low Low Low 
Aarnivala et al, 20176  Low Low Expert rating with Argenta classification Low Low Low 
Schaaf et al, 20107  Low Low Anthropometry Low Low Low 
Skolnick et al, 20148  Low Low Nonlinear measures from 3D photographs Unclear Low Low 
Porras et al, 201922  Low Low CT Low Low Low 
de Jong et al, 202023  Unclear Low CT Low Low Low 
Barbero-García et al, 202011  High Low CT/MRI Low Low High 
Barbero-García et al, 201818  Low Low CT/MRI Low Low Low 
Barbero-García et al, 202114  Low Low Target-based coded markers on a cap High High High 
Meulstee et al, 20179  Low Low CT Low Low Low 
Tu et al, 201924  Low Low CT Low Lowb Low 
Atmosukarto et al, 201010  Low Low Expert rating Low Low Low 
Author, YearDomain 1Domain 2Domain 3Domain 4Overall
Q1Q2Q3ResultQ1Q2ResultsReference StandardQ2Q3ResultQ1Q2Q3Q4Result
Wilbrand et al, 201125  Low Low Expert measurement Low Low Low 
Wu et al, 202012  Low Low CT Low Low Low 
Callejas Pastor et al, 202019  Unclear Low 3D cranial scanning Low Low Low 
Schaaf H et al, 201026  Low Low Anthropometry Low Low Low 
Lopes Alho et al, 202016  High High None High High High 
Agarwal et al, 201820  Low Low Professionally classified images Low Lowa Low 
Bookland et al, 202127  Low Low Professionally classified images/optical scan-derived craniometric measurements Low Low Low 
Geisler et al, 202128  Low Low Professionally classified images Low Low Low 
Hutchison et al, 200521  Low Low Flexicurve ruler Low Low Low 
van Adrichem et al, 200813  Low Low 3D CT Low Low Low 
Nahles et al, 201815  Low Low Anthropometry Low Low Low 
Aarnivala et al, 20176  Low Low Expert rating with Argenta classification Low Low Low 
Schaaf et al, 20107  Low Low Anthropometry Low Low Low 
Skolnick et al, 20148  Low Low Nonlinear measures from 3D photographs Unclear Low Low 
Porras et al, 201922  Low Low CT Low Low Low 
de Jong et al, 202023  Unclear Low CT Low Low Low 
Barbero-García et al, 202011  High Low CT/MRI Low Low High 
Barbero-García et al, 201818  Low Low CT/MRI Low Low Low 
Barbero-García et al, 202114  Low Low Target-based coded markers on a cap High High High 
Meulstee et al, 20179  Low Low CT Low Low Low 
Tu et al, 201924  Low Low CT Low Lowb Low 
Atmosukarto et al, 201010  Low Low Expert rating Low Low Low 

Y, yes; U, unclear; N, no.

a

Machine-learning database did not use “real patients.”

b

Healthy CT controls were used to construct a normative shape multiatlas. It is unclear if craniosynostosis patients were CT-confirmed, but they were status preop.

Included studies were published between 2005 and 2021. The authors of included studies reported on range of outcomes including accuracy, variability, time-to-diagnosis, comparison with gold-standard computed tomography (CT), and patient comfort; reported outcomes varied significantly by study. Patient age varied from 3 months to 12 years of age. Sample size for clinical studies ranged from 2 to 339. Six studies used preassembled data sets of cranial head images or models to evaluate performance. Ten studies reported sex, 5 of which reported >50% female participants.

Every modality included in this study aims to quantify cranial asymmetry, but all take different approaches (Fig 2, Table 2). Anthropometry uses calipers and measuring tapes to take assorted measurements of the child’s head. These measurements are then used to calculate cranial asymmetry indices (cranial vault asymmetry index [CVAI], cephalic index [CI], oblique cranial length ratio, etc), which can inform the user of the presence and severity of cranial deformity. PCM uses a thermoplastic band wrapped around the child’s head to make a mold that can be traced onto a paper. This trace can then be used to take measurements and calculate asymmetry indices without the child moving around. Digital photography commonly uses a photo taken from a birds-eye-view perspective of the child’s head. Measurements can then be taken directly off the photo with manual landmark selection, but new automated methods exist that take automatically calculated asymmetry indices (Fig 3). New digital photography techniques can use smartphones to take photos, and this ease of use translates to new three-dimensional (3D) photogrammetric techniques. 3D photogrammetry uses either the slow-motion features on modern smartphones or multicamera stationary imaging setups. In this method, a 3D image is pieced together from many still photos taken from different perspectives; cranial asymmetry can then be evaluated using 2D anthropometric indices or more complicated 3D indices. Finally, 3D laser scanning uses a stationary or handheld scanner to create a 3D representation of the infant’s head. For all image-based modalities (ie, photographic, photogrammetric, laser scanning), an infant’s hair presents an important confounder; to minimize its effect, infants are fitted with a nylon cap (with reflective dots in the case of laser scanning).

FIGURE 3

Processing pathway for automated digital photography measurement. A, Top view of an infant’s head affected by positional plagiocephaly. B, Overlay of a healthy head shape (dotted line) with digitally rendered cranial measurements used to calculate standard asymmetry indices (C).

FIGURE 3

Processing pathway for automated digital photography measurement. A, Top view of an infant’s head affected by positional plagiocephaly. B, Overlay of a healthy head shape (dotted line) with digitally rendered cranial measurements used to calculate standard asymmetry indices (C).

Close modal

A common theme across all modalities was the ability to characterize head shape. Specifically, quantifiable analysis of skull deformity was highlighted as an important trait in new diagnostic modalities. Shape analysis was divided between 2D methods (anthropometry, PCM, digital photography) and 3D methods (laser scanning, photogrammetry). 2D and 3D methods are distinguished by their speed and ability to evaluate multiple imaging planes. Measurements taken from these methods are calculated into indices, such as CI or CVAI, and used to evaluate the degree of deformity. 3D methods allow the development of more complex indices looking at deformity in multiple planes, but in some instances were found to retain the use of 2D indices.68  Deformational cranial growth is not restricted to the 2D plane, and in some cases may be more accurately diagnosed by 3D methods. Meulstee et al implemented principal component analysis to determine the mean cranial shape in the normal population and subsequently distinguish deviations from “normal” with 3D reconstructions obtained from photogrammetry.9  Atmosukarto et al demonstrated that their novel 3D-based plagiocephaly posterior severity scores were more capable of discriminating plagiocephalic and normal head shapes than 2D measurments.10  By contrast, Skolnick et al used nonlinear measures of cranial asymmetry from 3D photographs to identify the most capable linear measure, whereas Barbero-Garcia et al evaluated the performance of a 3D-photogrammetric method using automatically derived linear (2D) measurements.8,11  Wu et al (anthropometry) and Van Adrichem et al (PCM) demonstrated that 2D methods were capable of returning results that correlated with CT-derived measurements (P > .05 for both studies).12,13  Barbero-Garcia et al demonstrated an accuracy with tolerances <1.5 mm when comparing 3D-photogrammetry (smartphone-based) to 3D-CT.11  A later study by Barbero-Garcia et al combined machine learning-based facial analysis with their 3D-hotogrammetry system to improve cranial measurement in dynamic infants and capture facial asymmetries associated with cranial deformities; the authors concluded that eyes were the most consistently identifiable facial landmarks, although 3D coordinates of facial landmarks were not accurately obtained in 52.9% of cases.14  Because of equipment cost and the technical skill required to operate the system, 3D photogrammetry and laser scanning are generally centralized at a specialized location.13,1517  Laser scanning and 3D photogrammetry share key characteristics: quick capture times; precise, quantitative data collection; and an ability to track changes in head shape over time.17  Nonetheless, Nahles et al found that 3D laser scanning took, on average, >3 times longer than an anthropometric measurement.15  3D photogrammetry was shown by Aarnivala et al to not be affected by the age of the infant.6  Newer technology-based shape analysis methods were highlighted by high accuracy and improved analysis times: Barbero-Garcia presented a 3D-photogrammetric method that returned a diagnostic result in <5 minutes.7,9,11,18  One significant consideration in the clinical evaluation of head shape is the differentiation between positional plagiocephaly and craniosynostosis. Although Tu et al, Meulstee et al, Porras et al, Agarwal et al, and de Jong et al all demonstrated that their tools could reliably differentiate varying forms of synostosis from a healthy control, and several of the included studies were able to distinguish positional deformities from a healthy control, this review returned no studies that reported on diagnostic differentiation between positional and synostotic deformities.9,16,1924 

Given that a child may see different physicians at standard “well-visits” during the first few months of life, high inter- and intrarater reliability of cranial measurements are important to accurately monitor deformity progression. Anatomic landmarks used in different deformity indices were found to have varying intrarater reliability.8  Using 3D photogrammetry, Skolnick et al demonstrated that the contralateral frontozygomaticus to eurion measurement was the linear measure that best correlated with overall cranial asymmetry (r ≥ 0.90).8  Anthropometry and PCM have been shown to return acceptable interrater reliability.13,25  When implemented with a well-defined protocol, an anthropometric study led by Wilbrand et al returned intra- and interobserver variabilities of 0.03% (<1.131 mm2) and 0.5% (<0.182 mm2), respectively, with 2 mm of overall measurement variance.25  Van Adrichem et al reported no statistical difference (P > .05) between a PCM ring off the head and CT-based skull measurements with a plagiocephalometric method.13  Schaaf et al’s study concerning a digital photographic method reported interobserver correlation coefficients of 0.982 and 0.946 for CI and CVAI, respectively.26  In a distinct study by Schaaf et al, multicamera 3D photogrammetric methods (operated by trained specialists in a centralized location) returned an intraclass correlation coefficient of 0.97 for plagiocephaly and 0.98 for brachycephaly.7  More accessible to front-line physicians, 2 studies by Barbero-Garcia et al evaluating smartphone-based 3D photogrammetric methods have returned accurate measurements, with SD below 1.4 mm with a 99% confidence and differences in means (for intra- and interuser tests) below 1 mm with a 95% confidence interval.11,18 

Given the ubiquity of smartphones in the pockets of modern physicians, this review returned very few studies embracing the shift toward mobile optimization of previously specialized tools. Older digital photographic methods requiring dedicated handheld cameras and manual cranial measurements are now superseded by cellphone cameras for use in digital photographic diagnostic modalities.11,16  The network connectivity that new smartphones can leverage opens the doors for cloud computing, where the processing of a digital image is sent off to a server that returns a diagnostic output.11  As evidenced by Barbero-Garcia et al in 2019 and 2020, the implementation of native, slow-motion filming in smartphones is capable of automatic, handheld, and accessible photogrammetric 3D modeling.11,18  Although smartphone-based 3D-photogrammetric methods take longer to scan (<5 minutes) compared with specialized multicamera setups (<1 second), they are significantly more accessible and user friendly, with no requirement for specially trained users.7,9,11,18 

Anthropometry and PCM, as well as early digital and 3D photography, were limited by their dependence on manual determination of cranial landmarks, introducing errors and increasing variability between examiners. Technological advancement has provided a solution for this quandary: automated diagnosis. Automated diagnosis saw a 762% increase in the literature since 2017, with 11 out of 13 studies published since 2017 including a form of automation. Increases in computational ability and the adoption of AI by the medical community have proved promising for automated digital photographic assessment. Since 2017, Agarwal et al report a testing accuracy of 84.12% for a machine-learning model that identifies craniosynostosis; Porras et al described an algorithm which detected craniosynostosis automatically with 94.74% sensitivity and 96.02% specificity.20,22  It further correctly identified the fused sutures with 99.51% sensitivity and 99.13% specificity.22  de Jong et al also reported a deep learning algorithm for classifying craniosynostosis in which 195 out of 196 (99.5%) stereophotographs were correctly diagnosed.23  Bookland et al described their cranial shape classification software, which had an accuracy of 93.3% (95% confidence interval 86.8%–98.8%; P < .001), with a sensitivity of 92.0% and specificity of 94.3%.27  Geisler et al developed a convolutional neural network to classify synostosis; overall testing accuracy for their model was 90.6%, with higher sensitivity and precision when identifying metopic (100%, 100%) and sagittal (93.3%, 100%) synostosis compared with unicoronal synostosis (66.7%, 100%).28  Finally, Tu et al evaluated a support vector machine classifier that obtained a diagnostic accuracy of 91.03% for craniosynostosis, compared with 78.21% when applying methods that used head circumference/CI.24  Although diagnostic times for nonautomated methods were not reported in the literature, 3D scans taken with a smartphone returned a diagnostic result in as little as 2 minutes without sensitivity to the user.11  Notably, all the included automated diagnosis studies highlighted the need for access to larger training data sets as the most significant barrier to improved performance.16,19,20 

Recent technological advancements have allowed for the development of diagnostic tools to support pediatricians in monitoring head shape that are fast, accurate, and reliable. Shape analysis, smartphone integration, and automated diagnosis enhanced by AI have all seen considerable development in recent years, with promising potential for further improvement. Pediatricians manage large patient loads, and standard clinical and visual examination techniques lack the ability to accurately monitor changes in head shape. Subtle changes in the deformity over time can help inform treatment and referral decisions, so reliable and consistent monitoring is crucial to ensure positive neurologic and aesthetic outcomes.2931  The result of extrinsic forces on the infant’s skull in utero or early in life, cranial deformities will often normalize during the first few months of life with appropriate treatment.32,33  Increased (monitored) tummy time and physiotherapy are typically prescribed to address external causes such as torticollis; advanced cases may benefit from helmet-molding therapy.32,34,35  The effect of these treatments on head development can help distinguish deformational plagiocephaly from true craniosynostosis, and is valuable information for both pediatricians and the specialized craniofacial teams that treat true craniosynostosis. The gold standard diagnostic confirmation for suspected craniosynostosis is cranial 3D-CT.3639  It has been reported that outpatient CT exams (ie, those ordered by noncraniofacial specialists) on patients with cranial deformities return negative findings in 75% to 80% of cases.40  Specialists deciding whether patients with unclear clinical presentation should undergo CT would benefit from a tool providing detailed and quantitative history of head shape progression.

Accurate, accessible, and quantitative measurement of head shape is the cornerstone of early cranial deformity screening and eventual monitoring. A range of diagnostic modalities meeting these criteria were reported in the literature: anthropometry, PCM, digital photography, and 3D photogrammetry. 3D laser scanning was also included in the review, but was found to offer no advantages over much cheaper anthropometric techniques; Nahles et al recommended that, if available, 3D photogrammetry should be used until such time as developments in laser scanning techniques reduce the scanning time and increase the consistency of results.15  Although capable of returning accurate results, manual anthropometric techniques (including PCM) have been overtaken by newer modalities taking advantage of technological developments that automate and speed up diagnosis to support pediatricians in the clinic.41  In particular, smartphone-based digital photographic techniques for shape analysis have a promising future. Digital photos can leverage a modern smartphone’s network connectivity to access cloud servers for fast storage and retrieval, allowing physicians to directly compare quantitative (calculated asymmetry index) and subjective (visual head exam) data across multiple patient visits. Although smartphone cameras have lower radiometric accuracy than single lens reflex cameras, their ubiquity and portability more than make up for losses in image quality.42  Despite only being used for head shape analysis since 2019, smartphone cameras have previously been successfully implemented across many medical fields, including plastic surgery for 3D facial scanning, monitoring microvascular responses to physiologic provocations in the skin, and augmented reality microsurgical planning for lymphovenous anastomosis.4345  The use of slow-motion video-recording on smartphones enables low-cost 3D-photogrametric modeling that is insensitive to use by nonmedical users while still providing results comparable to CT.11  These methods are highly automated and can return results to the physician in as little as 4 minutes.11  Plain photographs from smartphones can also leverage automatic measurement extraction for fast quantitative results.16,20  Accounting for motion is a crucial consideration when dealing with a patient population that has difficulty sitting still, and it will play an important role in the efficacy of future techniques. Particularly in smartphone-rendered 3D models where capture speeds are not instantaneous, innovative and computationally efficient ways to deal with infant motion are required. Barbero-Garcia et al’s method accomplished motion-insensitivity by automatically overlapping images using a coded cap and discarding any nonideal frames.11  It is important to emphasize that the case for smartphone-based diagnostic software is still in its infancy; although new developments are promising for future integration into clinical practice, more refinement is required for successful, widespread adoption. Current systems supporting the generic diagnosis of head deformity are highly performant, but translation of these tools for the diagnosis of craniosynostosis (a deformity distinguished primarily by internal pathology) will likely require significant databases and a helping hand from AI algorithms. When coupled with improved AI systems, mobile photographic methods have the potential to put automated diagnostic software on par with trained radiologists in physicians’ pockets.

The introduction of AI and automation into newer digital photographic solutions has the potential to drive down exam times and further enhances clinical viability.4,20  Model performance is dependent on the quality and size of the training set used to instruct it, and there are currently no sufficiently large, high-quality training databases for the modalities included in this review.19,20,2224  There have been promising attempts to artificially expand data sets using a generative adversarial network (a deep learning technique), which outputs data similar (but not the same) as the data input of the network. The method of multiple runs has also proved promising to maximize training on a small data set.46  Although similar techniques warrant more research, they do not obviate the need for large, annotated, craniofacial databases based on real patient data. Conversely, radiographic imaging has access to massive amounts of data. The implementation of the picture archiving and communication systems in the United States, Canada, and Europe has provided unparalleled access to clinical imaging data sets.47,48  The 2018–19 year alone saw the addition of 44.8 million imaging tests to the National Health Service picture archiving and communication system in England.49  Correspondingly, AI has seen greater success and implementation in supporting radiologists; a 2020 study introduced a new AI system for predicting breast cancer in screening mammography’s that outperformed all of the expert human readers involved.50  More relevantly, a machine-learning algorithm has been trained to distinguish between types of craniosynostosis on CT scans of patients and return the correct diagnosis, with a sensitivity of 92.3% and a specificity of 98.9%.51  The development of training sets for nonradiographic diagnostic modalities should increase our ability to develop deep learning and convolutional neural network-based frameworks for cranial deformity diagnosis. Additionally, expanded data sets will play a crucial role in establishing clinically relevant diagnostic cutoff values for deformational indices. Indicators of craniofacial deformity (CVAI, CI, oblique cranial length ratio, etc) are based on the 2D measurements found in anthropometric methods, but 3D models can be evaluated by novel multidimensional indicators of cranial deformity.6,810  To fully leverage the diagnostic ability of AI systems, clinical indicators for cranial deformity need clearly defined diagnostic cutoffs for measurement values.

There are several limitations found in our systematic review. Most notably, there is significant heterogeneity in the presentation and quality of data in studies presenting diagnostic tools, whether as a proof of concept or a validation study. A lack of common performance reporting practices across different modalities creates a challenge in translating results for comparison. Certain modalities, such as 3D laser scanning, were also found to be significantly underreported, despite being highly cited in the literature. A crucial variable in the evaluation of front-line technology integration is time; most of the studies in this review did not report the time required for a given examination. Paired studies also usually neglected to compare patient (infant and parent) preferences.

Pediatricians have access to a comprehensive range of tools when evaluating for and monitoring potential craniofacial deformities. Although limited by the reporting practices seen in the literature, substantial evidence exists to support the use of these tools by pediatricians for clinical monitoring of cranial deformity in infants. There are a multitude of potential diagnostic options, and selection will likely be made on a case-by-case basis depending on resource availability. Regardless of the choice, successful diagnostic tools were all reliable, quantitative, and fast, with newer mobile solutions being increasingly cost-effective. Smartphone-based options, particularly those leveraging AI classification algorithms, hold great promise by placing significant diagnostic power within the pockets of physicians around the world. Particularly for those physicians without easy access to a tertiary care center and craniofacial specialists, future work optimizing mobile diagnostic tools has the potential to significantly improve diagnostic times and, consequently, patient outcomes.

FUNDING: No external funding.

Mr Watt conceptualized and designed the study, designed the data collection instruments, collected data, conducted the initial analyses, drafted the initial manuscript, and reviewed and revised the manuscript; Dr Zammit conceptualized and designed the study, designed the data collection instruments, collected data, and reviewed and revised the manuscript; Drs Gilardino and Lee supervised data collection, reviewed and revised the manuscript, and critically reviewed the manuscript for important intellectual content; and all authors approved the final manuscript as submitted and agree to be accountable for all aspects of the work.

2D

two-dimensional

3D

three-dimensional

AI

artificial intelligence

CT

computed tomography

CI

cephalic index

CVAI

cranial vault asymmetry index

PCM

plagiocephalometry

1
Di Rocco
F
,
Ble
V
,
Beuriat
PA
,
Szathmari
A
,
Lohkamp
LN
,
Mottolese
C
.
Prevalence and severity of positional plagiocephaly in children and adolescents
.
Acta Neurochir (Wien)
.
2019
;
161
(
6
):
1095
1098
2
Kunz
F
,
Schweitzer
T
,
Große
S
, et al
.
Head orthosis therapy in positional plagiocephaly: longitudinal 3D-investigation of long-term outcomes, compared with untreated infants and with a control group
.
Eur J Orthod
.
2019
;
41
(
1
):
29
37
3
Pogliani
L
,
Mameli
C
,
Fabiano
V
,
Zuccotti
GV
.
Positional plagiocephaly: what the pediatrician needs to know. A review
.
Childs Nerv Syst
.
2011
;
27
(
11
):
1867
1876
4
Ahuja
AS
.
The impact of artificial intelligence in medicine on the future role of the physician
.
PeerJ
.
2019
;
7
:
e7702
5
Whiting
PF
,
Rutjes
AW
,
Westwood
ME
, et al.
QUADAS-2 Group
.
QUADAS-2: a revised tool for the quality assessment of diagnostic accuracy studies
.
Ann Intern Med
.
2011
;
155
(
8
):
529
536
6
Aarnivala
H
,
Vuollo
V
,
Heikkinen
T
, et al
.
Accuracy of measurements used to quantify cranial asymmetry in deformational plagiocephaly
.
J Craniomaxillofac Surg
.
2017
;
45
(
8
):
1349
1356
7
Schaaf
H
,
Pons-Kuehnemann
J
,
Malik
CY
, et al
.
Accuracy of three-dimensional photogrammetric images in non-synostotic cranial deformities
.
Neuropediatrics
.
2010
;
41
(
1
):
24
29
8
Skolnick
GB
,
Naidoo
SD
,
Patel
KB
,
Woo
AS
.
Analysis of digital measures of cranial vault asymmetry for assessment of plagiocephaly
.
J Craniofac Surg
.
2014
;
25
(
4
):
1178
1182
9
Meulstee
JW
,
Verhamme
LM
,
Borstlap
WA
, et al
.
A new method for three-dimensional evaluation of the cranial shape and the automatic identification of craniosynostosis using 3D stereophotogrammetry
.
Int J Oral Maxillofac Surg
.
2017
;
46
(
7
):
819
826
10
Atmosukarto
I
,
Shapiro
LG
,
Starr
JR
, et al
.
Three-dimensional head shape quantification for infants with and without deformational plagiocephaly
.
Cleft Palate Craniofac J
.
2010
;
47
(
4
):
368
377
11
Barbero-García
I
,
Lerma
JL
,
Mora-Navarro
G
.
Fully automatic smartphone-based photogrammetric 3D modelling of infant’s heads for cranial deformation analysis
.
ISPRS J Photogramm Remote Sens
.
2020
;
166
:
268
277
12
Wu
GT
,
Audlin
JR
,
Grewal
JS
,
Tatum
SA
.
Comparing caliper versus computed tomography measurements of cranial dimensions in children
.
Laryngoscope
.
2021
;
131
(
4
):
773
775
13
van Adrichem
LN
,
van Vlimmeren
LA
,
Cadanová
D
, et al
.
Validation of a simple method for measuring cranial deformities (plagiocephalometry)
.
J Craniofac Surg
.
2008
;
19
(
1
):
15
21
14
Barbero-García
I
,
Pierdicca
R
,
Paolanti
M
,
Felicetti
A
,
Lerma
JL
.
Combining machine learning and close-range photogrammetry for infant’s head 3D measurement: a smartphone-based solution
.
Measurement
.
2021
;
182
:
109686
15
Nahles
S
,
Klein
M
,
Yacoub
A
,
Neyer
J
.
Evaluation of positional plagiocephaly: conventional anthropometric measurement versus laser scanning method
.
J Craniomaxillofac Surg
.
2018
;
46
(
1
):
11
21
16
Lopes Alho
EJ
,
Rondinoni
C
,
Furokawa
FO
,
Monaco
BA
.
Computer-assisted craniometric evaluation for diagnosis and follow-up of craniofacial asymmetries: SymMetric v. 1.0
.
Childs Nerv Syst
.
2020
;
36
(
6
):
1255
1261
17
Purnell
CA
,
Benz
AW
,
Gosain
AK
.
Assessment of head shape by craniofacial teams: structuring practice parameters to optimize efficiency
.
J Craniofac Surg
.
2015
;
26
(
6
):
1808
1811
18
Barbero-García
I
,
Lerma
JL
,
Miranda
P
,
Marqués-Mateu
Á
.
Smartphone-based photogrammetric 3D modelling assessment by comparison with radiological medical imaging for cranial deformation analysis
.
Measurement
.
2018
;
131
:
372
379
19
Callejas Pastor
CA
,
Jung
IY
,
Seo
S
,
Kwon
SB
,
Ku
Y
,
Choi
J
.
Two-Dimensional Image-Based Screening Tool for Infants with Positional Cranial Deformities: A Machine Learning Approach
.
Diagnostics (Basel)
.
2020
;
10
(
7
):
495
20
Agarwal
S
,
Hallac
RR
,
Mishra
R
,
Li
C
,
Daescu
O
,
Kane
A
.
Image based detection of craniofacial abnormalities using feature extraction by classical convolutional neural network
.
2018 IEEE 8th International Conference on Computational Advances in Bio and Medical Sciences (ICCABS)
.
Oct 2018
21
Hutchison
BL
,
Hutchison
LA
,
Thompson
JM
,
Mitchell
EA
.
Quantification of plagiocephaly and brachycephaly in infants using a digital photographic technique
.
Cleft Palate Craniofac J
.
2005
;
42
(
5
):
539
547
22
Porras
AR
,
Tu
L
,
Tsering
D
, et al
.
Quantification of head shape from three-dimensional photography for presurgical and postsurgical evaluation of craniosynostosis
.
Plast Reconstr Surg
.
2019
;
144
(
6
):
1051e
1060e
23
de Jong
G
,
Bijlsma
E
,
Meulstee
J
, et al
.
Combining deep learning with 3D stereophotogrammetry for craniosynostosis diagnosis
.
Scientific Reports
.
2020
;
10
(
1
):
15346
24
Tu
L
,
Porras
AR
,
Oh
A
, et al
.
Quantitative evaluation of local head malformations from three-dimensional photography: application to craniosynostosis
.
Proc SPIE Int Soc Opt Eng
.
Feb
2019
25
Wilbrand
JF
,
Wilbrand
M
,
Pons-Kuehnemann
J
, et al
.
Value and reliability of anthropometric measurements of cranial deformity in early childhood
.
J Craniomaxillofac Surg
.
2011
;
39
(
1
):
24
29
26
Schaaf
H
,
Wilbrand
JF
,
Boedeker
RH
,
Howaldt
HP
.
Accuracy of photographic assessment compared with standard anthropometric measurements in nonsynostotic cranial deformities
.
Cleft Palate Craniofac J
.
2010
;
47
(
5
):
447
453
27
Bookland
MJ
,
Ahn
ES
,
Stoltz
P
,
Martin
JE
.
Image processing and machine learning for telehealth craniosynostosis screening in newborns
.
J Neurosurg Pediatr
.
2021
;
27
(
5
):
581
588
.
28
Geisler
EL
,
Agarwal
S
,
Hallac
RR
,
Daescu
O
,
Kane
AA
.
A role for artificial intelligence in the classification of craniofacial anomalies
.
J Craniofac Surg
.
2021
;
32
(
3
):
967
969
29
Di Chiara
A
,
La Rosa
E
,
Ramieri
V
,
Vellone
V
,
Cascone
P
.
Treatment of deformational plagiocephaly with physiotherapy
.
J Craniofac Surg
.
2019
;
30
(
7
):
2008
2013
30
Collett
B
,
Breiger
D
,
King
D
,
Cunningham
M
,
Speltz
M
.
Neurodevelopmental implications of “deformational” plagiocephaly
.
J Dev Behav Pediatr
.
2005
;
26
(
5
):
379
389
31
Nagaraja
S
,
Anslow
P
,
Winter
B
.
Craniosynostosis
.
Clinical Radiology
.
2013
;
68
(
3
):
284
292
32
Beuriat
PA
,
Szathmari
A
,
Di Rocco
F
,
Mottolese
C
.
Deformational plagiocephaly: state of the art and review of the literature
.
Neurochirurgie
.
2019
;
65
(
5
):
322
329
33
Jung
BK
,
Yun
IS
.
Diagnosis and treatment of positional plagiocephaly
.
Arch Craniofac Surg
.
2020
;
21
(
2
):
80
86
34
Kuo
AA
,
Tritasavit
S
,
Graham
JM
Jr
.
Congenital muscular torticollis and positional plagiocephaly
.
Pediatr Rev
.
2014
;
35
(
2
):
79
87
;
Quiz 87
35
Çevik
S
,
Işık
S
,
Özkılıç
A
.
The role of age on helmet therapy in deformational plagiocephaly and asymmetric brachycephaly
.
Childs Nerv Syst
.
2020
;
36
(
4
):
803
810
36
Saarikko
A
,
Mellanen
E
,
Kuusela
L
, et al
.
Comparison of Black Bone MRI and 3D-CT in the preoperative evaluation of patients with craniosynostosis
.
J Plast Reconstr Aesthet Surg
.
2020
;
73
(
4
):
723
731
37
Ginat
DT
,
Lam
D
,
Kuhn
AS
,
Reid
R
.
CT imaging findings after craniosynostosis reconstructive surgery
.
Pediatr Neurosurg
.
2018
;
53
(
4
):
215
221
38
Montoya
JC
,
Eckel
LJ
,
DeLone
DR
, et al
.
Low-dose CT for craniosynostosis: preserving diagnostic benefit with substantial radiation dose reduction
.
AJNR Am J Neuroradiol
.
2017
;
38
(
4
):
672
677
39
Massimi
L
,
Bianchi
F
,
Frassanito
P
,
Calandrelli
R
,
Tamburrini
G
,
Caldarelli
M
.
Imaging in craniosynostosis: when and what?
Childs Nerv Syst
.
2019
;
35
(
11
):
2055
2069
40
Simanovsky
N
,
Hiller
N
,
Koplewitz
B
,
Rozovsky
K
.
Effectiveness of ultrasonographic evaluation of the cranial sutures in children with suspected craniosynostosis
.
Eur Radiol
.
2009
;
19
(
3
):
687
692
41
King
ALS
,
Pádua
MK
,
Gonçalves
LL
,
Santana de Souza Martins
A
,
Nardi
AE
.
Smartphone use by health professionals: a review
.
Digit Health
.
2020
;
6
:
2055207620966860
42
Barbero-García
I
,
Cabrelles
M
,
Lerma
JL
,
Marqués-Mateu
Á
.
Smartphone-based close-range photogrammetric assessment of spherical objects
.
The Photogrammetric Record
.
2018
;
33
(
162
):
283
299
43
Rudy
HL
,
Wake
N
,
Yee
J
,
Garfein
ES
,
Tepper
OM
.
Three-dimensional facial scanning at the fingertips of patients and surgeons: accuracy and precision testing of iphone X three-dimensional scanner
.
Plast Reconstr Surg
.
2020
;
146
(
6
):
1407
1417
44
Tesselaar
E
,
Farnebo
S
.
Objective assessment of skin microcirculation using a smartphone camera
.
Skin Res Technol
.
2020
;
27
(
2
):
138
144
45
Pereira
N
,
Venegas
J
.
Augmented reality microsurgical planning with a smartphone for lymphovenous anastomosis
.
Plast Reconstr Surg
.
2019
;
144
(
5
):
955e
956e
46
Shaikhina
T
,
Khovanova
NA
.
Handling limited datasets with neural networks in medical applications: a small-data approach
.
Artif Intell Med
.
2017
;
75
:
51
63
47
Kim
DH
,
MacKinnon
T
.
Artificial intelligence in fracture detection: transfer learning from deep convolutional neural networks
.
Clin Radiol
.
2018
;
73
(
5
):
439
445
48
Hosny
A
,
Parmar
C
,
Quackenbush
J
,
Schwartz
LH
,
Aerts
HJWL
.
Artificial intelligence in radiology
.
Nat Rev Cancer
.
2018
;
18
(
8
):
500
510
49
National Health Service England
.
Diagnostic Imaging Dataset Annual Statistical Release 2018/19
.
2019
.
Available at: https://www.england.nhs.uk/statistics/. Accessed January 2, 2022
50
McKinney
SM
,
Sieniek
M
,
Godbole
V
, et al
.
International evaluation of an AI system for breast cancer screening
.
Nature
.
2020
;
577
:
89
94
51
Mendoza
CS
,
Safdar
N
,
Okada
K
,
Myers
E
,
Rogers
GF
,
Linguraru
MG
.
Personalized assessment of craniosynostosis via statistical shape modeling
.
Med Image Anal
.
2014
;
18
(
4
):
635
646

Competing Interests

FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.

POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no conflicts of interest to disclose.