3D morphometric quantification of maxillae and defects for patients with unilateral cleft palate via deep learning-based CBCT image auto-segmentation

Xiaoyu Wang; Matthew Pastewait; Tai-Hsien Wu; Chunfeng Lian; Beatriz Tejera; Yan-Ting Lee; Feng-Chang Lin; Li Wang; Dinggang Shen; Song Li; Ching-Chang Ko

doi:10.1111/ocr.12482

. Author manuscript; available in PMC: 2022 Dec 1.

Published in final edited form as: Orthod Craniofac Res. 2021 Mar 25;24(Suppl 2):108–116. doi: 10.1111/ocr.12482

3D morphometric quantification of maxillae and defects for patients with unilateral cleft palate via deep learning-based CBCT image auto-segmentation

Xiaoyu Wang ^a,^b, Matthew Pastewait ^c, Tai-Hsien Wu ^d, Chunfeng Lian ^e, Beatriz Tejera ^f, Yan-Ting Lee ^g, Feng-Chang Lin ^h, Li Wang ⁱ, Dinggang Shen ^j, Song Li ^k, Ching-Chang Ko ^l

^a.Orthodontist, Department of Orthodontics, Beijing Stomatological Hospital, Capital Medical University. No.4 Tiantan Xili, Dongcheng District, Beijing, P. R. China, 100050

^b.Orthodontist, Department of Stomatology, Beijing Tian Tan Hospital, Capital Medical University. No.119 South Fourth Ring West Road, Fengtai District, Beijing, P. R. China, 100070

^c.Orthodontist, United States Air Force, Kadena AB, Japan

^d.Postdoctoral Scholar, Division of Orthodontics, College of Dentistry, the Ohio State University, Columbus, OH, 43210, USA

^e.Associate Professor, School of Mathematics and Statistics, Xi’an Jiaotong University, Shaanxi 710049, China

^f.Resident, Orthodontics, Nova Southeastern University, Ft. Lauderdale, Florida, 33328 USA

^g.Research Assistant, Oral and Craniofacial Health Sciences Research, Adam School of Dentistry, University of North Carolina, CB #7455, Chapel Hill, NC, 27599, USA

^h.Associate Professor, Department of Biostatistics, University of North Carolina, Chapel Hill, NC, USA

^i.Assistant Professor, Department of Radiology and Biomedical Research Imaging Center, University of North Carolina, CB #7513, Chapel Hill, NC, 27599, USA

^j.Professor, School of Biomedical Engineering, ShanghaiTech University, Shanghai, China; Department of Research and Development, Shanghai United Imaging Intelligence Co., Ltd., Shanghai, China; Department of Artificial Intelligence, Korea University, Seoul, Republic of Korea

^k.Professor, Department of Orthodontics, Beijing Stomatological Hospital, Capital Medical University. No.4 Tiantan Xili, Dongcheng District, Beijing, P. R. China, 100050

^l.Professor, Division of Orthodontics, College of Dentistry, the Ohio State University, Columbus, OH, 43210, USA

Authors contributions

Xiaoyu Wang: contributed to image acquisition, elaboration, analysis of results, and writing of the article.

Matthew Pastewait: contributed to elaboration and writing of the article.

Tai-Hsien Wu: contributed to writing, data analysis, development of deep learning model, and supervising overall project including planning and design.

Chunfeng Lian: contributed to development of deep learning model.

Beatriz Tejera: contributed to elaboration and writing of the article.

Yan-Ting Lee: contributed to analysis of results.

Feng-Chang Lin: contributed to analysis of results.

Li Wang: contributed to development of deep learning model.

Dinggang Shen: contributed to supervising overall project including planning and design.

Song Li: contributed to supervising overall project including planning and design.

Ching-Chang Ko: contributed to supervising overall project including planning and design.

^✉

Corresponding authors: 1.Tai-Hsien Wu, wu.5084@osu.edu, Division of Orthodontics, College of Dentistry, the Ohio State University, Postle Hall, Rm 4088A, 305 W. 12th Avenue, Columbus, OH 43210; 2.Song Li, lisong@ccmu.edu.cn, Beijing Stomatological Hospital, Capital Medical University – Department of Orthodontics, No.4 Tiantan Xili, Dongcheng District, Beijing, P. R. China, 100050

PMCID: PMC8435046 NIHMSID: NIHMS1684142 PMID: 33711187

Abstract

Objective:

This study aimed to quantify the 3D asymmetry of the maxilla in patients with unilateral cleft lip and palate (UCP) and investigate the defect factors responsible for the variability of the maxilla on the cleft side using a deep learning-based CBCT image segmentation protocol.

Setting and Sample Population:

CBCT images of 60 patients with UCP were acquired. The samples in this study consisted of 39 males and 21 females, with a mean age of 11.52 years (SD=3.27 years; range of 8–18 years).

Materials and Methods:

The deep learning-based protocol was used to segment the maxilla and defect initially, followed by manual refinement. Paired t-tests were performed to characterize the maxillary asymmetry. A multiple linear regression was carried out to investigate the relationship between the defect parameters and those of the cleft side of the maxilla.

Results:

The cleft side of the maxilla demonstrated a significant decrease in maxillary volume and length as well as alveolar length, anterior width, posterior width, anterior height, and posterior height. A significant increase in maxillary anterior width was demonstrated on the cleft side of the maxilla. There was a close relationship between the defect parameters and those of the cleft side of the maxilla.

Conclusions:

Based on the 3D volumetric segmentations, significant hypoplasia of the maxilla on the cleft side existed in the pyriform aperture and alveolar crest area near the defect. The defect structures appeared to contribute to the variability of the maxilla on the cleft side.

INTRODUCTION

Unilateral cleft lip and palate (UCP) is a common congenital maxillofacial hypoplasia that exhibits a high incidence. Data for 7.5 million births from the international perinatal database of typical oral clefts (IPDTOC) showed that the prevalence of cleft lip and palate was 6.64 per 10,000 births worldwide.¹ The resultant deformity, located in the maxilla and other midfacial areas, is characterized by the incomplete formation of the lip, alveolar crest, hard palate, and soft palate.² It often results in some combination of feeding, deglutition, speaking, hearing, and/or psychological problems.³

Facial morphology plays a crucial role in mental health and social recognition. The fundamental goal of reconstruction is to restore the facial symmetry three-dimensionally (3D) to within a clinically acceptable range of the general population.⁴ Thus, there is an increasing need to quantitatively measure and accurately describe the extent of facial asymmetry. Many studies have attempted to investigate the morphological features of the maxilla in UCP patients^4–7. There is still no consistent conclusion in terms of whether significant asymmetry exists in the “deeper” maxilla. Several studies demonstrated that the maxillary asymmetry is confined to “local” regions of the cleft,^5,6 while others stated that the asymmetry also involves “deeper” regions of the maxilla.^4,7 From an etiological point of view, no single factor has been identified as a cause of the maxillary asymmetry. Several studies indicated that numerous etiological factors might contribute to this asymmetry, such as the location and extent of the defect, deviation of the nasal septum cartilage, and side effects of previous surgery.^8,9 Since the cleft is located in the maxilla, it is imperative to understand the morphology of the defect and the defect-related maxillary dysmorphology. Barbosa et al. revealed some clues indicating that the defect volume is related to the gap, arch, nasal and dental parameters to a certain degree.¹⁰ However, few studies focused on revealing the morphological relationship between the defect and affected maxillary half.

An essential step in the morphological analysis workflow is to obtain a virtual 3D model of the maxilla and defect. Although manual segmentation is considered the gold standard, it has the inherent drawback of being time-consuming, which affects its application in large-scale clinical studies. With the rapid advance in artificial intelligence (AI), several methods for auto-segmentation of the region of interest have been proposed in the literature.^11,12 Chang et al. proposed a model-based segmentation approach for a particular interest in the outer surface of the anterior wall of the maxilla. However, the thin bony structure surrounding the maxillary sinus is challenging to identify and segment from the adjacent maxillofacial bones.¹³ Compared with the auto-segmentation of the maxilla, the segmentation of the defect is more difficult. The defect has no specific boundary with the nasal and oral cavities, and its image intensity is similar to the soft tissue in the CBCT. Recently, Zhang et al. utilized 3D U-Net^14,15 which incorporated non-rigid volumetric registration to explore the discrepancy between complete and diseased maxillae.¹⁶ Using this algorithm, they obtained more accurate segmentations of the maxilla and cleft for UCP patients, compared to other methods. Despite the fact that these state-of-the-art AI algorithms still cannot be directly applied in a clinical study in a fully automatic manner, we can use them to obtain initial segmentation results, followed by manual post-processing refinement, to achieve clinically applicable segmentations on our dataset. In this way, the preparation time for a large sample size can be significantly reduced.

This study aimed to investigate the defect factors responsible for the variability of the maxilla on the cleft side by quantifying the 3D asymmetry of the maxilla in patients with UCP based on a deep learning-based CBCT image auto-segmentation method with manual refinement.

MATERIALS AND METHODS

Subjects

This retrospective study was performed under the approval of the institutional review board of Beijing Tian Tan Hospital, Capital Medical University (KY2017–072-01). The CBCT images of 60 subjects were selected from the Department of Orthodontics, Beijing Stomatological Hospital, Capital Medical University. All of the subjects had been diagnosed with non-syndromic UCP and received primary lip and palate repair. The subjects who had a history of alveolar bone grafts, orthodontic treatment, maxillofacial neoplasia, trauma, or orthognathic surgery were excluded. A total of 60 subjects were included in this study, consisting of 39 males and 21 females, with a mean age of 11.52 years (SD=3.27 years; range of 8–18 years), presenting with 41 left-side defects and 19 right-side defects.

All of these CBCT images were obtained during routine radiographic documentation prior to orthodontic treatment or surgical planning. All CBCT images were acquired on the NewTom VG scanner (QR srl, Verona, Italy) under a standard scanning protocol: 110 kV, 6.35 mA, 15×15 cm field of view, 0.250 mm slice thickness. The subjects were in an upright position for maximum intercuspation. The Frankfort plane was parallel to the floor.

Maxilla and Defect Segmentation

The 60 CBCT image sets were exported in DICOM format. To reduce the learning difficulty for the AI algorithm, the orientation of each CBCT image was adjusted to have every defect on the left side of the maxilla. We randomly selected 30 CBCT images and included them in Group 1. The remaining 30 images were assigned to Group 2. All the CBCT images in Group 1 were annotated manually and semi-automatically using ITK-SNAP¹⁷ (version 3.6.0; www.itksnap.org) by two trained specialists, serving as the gold standard. The manual segmentations from the two specialists had Cohen’s kappa statistic of 0.98, indicating our manual segmentations were in the high standard. For images in Group 1, the maxilla was manually segmented by tracing the boundaries where it articulated with the adjacent bony structures in the craniofacial region. The defect was manually segmented by following the contralateral shape to obtain a continuation of the alveolar ridge and hard palate.

The automatic segmentation protocol used in this study is 3D U-Net^14,15, which is a well-known deep learning neural network for biomedical image segmentation. The network architecture of 3D U-Net consists of three elements: contracting path, expansive path, and skip connections between the two paths. The contracting path is used to capture the context in the image, whereas the expansive path is used to recover local information based on deconvolution. The skip connections between the two paths help to fuse the local and global features, resulting in its outstanding performance in image segmentation.

A 3D U-Net model was trained and tested based on the 30 CBCT images in Group 1, where 24 images served as the training set, 3 images as the validation set, and the remaining 3 images as the test set. In the training set, each CBCT image generated 3,000 patches with a voxel size of $64 \times 64 \times 64$ with voxel spacing of 0.25 mm, which served as the real training samples. The amount of training data in terms of patches was much larger than the number of original images, which helped to overcome the limited number of samples in the biomedical field. The model was optimized to have minimal generalized Dice loss¹⁸ for 50 epochs using the Adam optimizer¹⁹. All processes were implemented using PyTorch²⁰, an open-source deep learning library developed by Facebook.

Each CBCT was split into several patches with the same size of training patches in the sliding window manner. The output was the patch containing the voxel label, and its size was the same as the input patch. Combining the output patches, we obtained the segmentation result of the CBCT. The accuracy of the segmentation model is evaluated by the Dice similarity coefficient (DSC), defined as follows:

D S C = \frac{2 | A \cap B |}{| A | + | B |}

where $| A |$ and $| B |$ represent the cardinalities of the learned and manual sets, and $| A \cap B |$ represents the intersection of the two sets. A value of 0 indicates no similarity, whereas a value of 1 indicates perfect agreement.

The trained 3D U-Net model was then used to predict the segmentation for the remaining 30 CBCT images in Group 2. An orthodontist further refined the predicted segmentation results following the automatic segmentation procedure. In addition, to eliminate the effect of the number of teeth on the volumetric measurements (especially for cases at different stages of dental development), all crowns were manually removed in all of the CBCT images (both Group 1 and Group 2). All 60 pairs of segmented maxilla and defect models were then used for morphometric and statistical analysis.

Description of Measurement

For 3D morphometric quantification, four midsagittal landmarks and 10 bilateral landmarks, as defined in Table 1 and illustrated in Figures 1A and 1B, were identified on the surface of the 3D segmented model and verified in the multiple planar reformat mode. Three reference planes (the Frankfort horizontal plane, midsagittal plane, and coronal plane) established a coordinate system (Table 1). The maxilla was separated by the midsagittal plane for measurements on the cleft and non-cleft sides. Eleven structural parameters of the maxilla (10 linear distances and 1 volume, including the maxilla and alveolar crest) were measured in horizontal, midsagittal, and coronal plane projections to characterize the maxillary asymmetry. In addition, four structural parameters of the defect (three linear distances and one volume) were measured, in order to investigate their correlation with those of the maxilla on the cleft side. All parameters are defined in Table 1, while various defect measurements are shown in Figure 1C.

Table 1:

3D cephalometric landmarks, reference planes and measurements of the maxilla and defect

Items	Definition
Landmarks
N	Intersection of internasal suture with nasofrontal suture
S	Midpoint of sella turcica
ANS	Most anterior point of anterior nasal spine
PNS	Most posterior point of posterior nasal spine
Po	Uppermost point on bony external auditory meatus
Sm	Superior most extent of the maxilla
Or	Lowest point on infraorbital edge
Lap	Most lateral points on the nasal aperture
J	Intersection of the outline of the tuberosity of the maxilla and zygomatic buttress
Mt	Posterior most extent of the maxillary tuberosity
Am	Posterior most extent of the anterior contour of the maxilla
Spc	Midpoint of labial alveolar crest of maxillary canine (No missing canine was observed in all subjects)
Spm	Midpoint of buccal alveolar crest of maxillary first molar
Aa	Most inferior anterior point of alveolar crest
Reference planes
Horizontal plane (FH plane)	Plane that passes through the bilateral Po and Or on the non-defect side
Midsagittal plane (MS plane)	Plane perpendicular to the FH plane passing through the N and S
Coronal plane (CR plane)	Plane perpendicular to the FH and MS plane passing through the N
Measurements
Maxillay lengh (L_max)	Sagittal distance from Am to Mt
Maxillary anterior width (AntW_max)	Transverse distance from Lap to the MS plane
Maxillary posterior width (PosW_max)	Transverse distance from J to the MS plane
Maxillary anterior height (AntH_max)	Vertical distance from Or to ANS
Maxillary posterior height (PosH_max)	Vertical distance from Sm to PNS
Maxillary volume (V_max)	Volume of the segmented individual maxilla
Alveolar lengh (L_alv)	Maximum sagittal distance from Aa to Mt
Alveolar anterior width (AntW_alv)	Transverse distance from Spc to the MS plane
Alveolar posterior width (PosW_alv)	Transverse distance from Spm to the MS plane
Alveolar anterior height (AntH_alv)	Vertical distance from Spc to ANS
Alveolar posterior height (PosH_alv)	Vertical distance from Spm to PNS
Defect length (L_def)	Maximum Sagittal distance of the defect
Defect width (W_def)	Maximum transverse distance of the defect
Defect height (H_def)	Maximum vertical distance of the defect
Defect volume (V_def)	Volume of the segmented defect

Open in a new tab

Figure 1: — The main landmarks on the 3D segmented model. A: Frontal view (cleft side and non-cleft side; landmarks on the non-cleft side are denoted by superscript ‘); B: Lateral view (cleft side); C: 3D segmentation measurements of the defect (L_def, W_def, and H_def indicate the defect length, width, and height, respectively).

The lengths, widths, and heights were measured by calculating the distances between the position (voxel coordinates) of landmarks, and the volumes of the maxilla and defect were measured based on the segmentation voxel counting. These measurements were carried out using ITK-SNAP.

Statistical Analysis

To assess intra-observer consistency and inter-observer reliability, landmark identifications and distance measurements were repeated two weeks after the first examination by two trained specialists. The intra-observer consistency, which is measured by the intra-class correlation coefficient (ICC), is greater than 0.90, confirming the consistency of the measurements. The inter-observer reliability, measured by the difference between the two observers, was tested by a paired t-test, having the p-value of 0.39, indicating a good agreement between two observers. The mean values of measurements were used for statistical analysis. The data are presented as mean values and standard deviations. The maxillary asymmetry between the cleft and non-cleft sides was compared using the paired samples t-test. A multiple linear regression was carried out to analyze the relationship between the parameters of the defect and those of the cleft side of the maxilla, with adjustments for the age and gender of subjects in the regression. The Bonferroni correction was used to control for the family-wise type-I error probability of multiple comparisons. All statistical analyses were performed using SPSS (Version 19.0; IBM Co., Armonk, NY, USA). The level of significance was set at P<0.05.

RESULTS

Maxilla and Defect Segmentation

The Dice similarity coefficients of the maxilla and defect between the manual and auto-segmentation samples were 0.92±0.01 and 0.77±0.06, respectively. Figure 2 shows the automatic (the first row) and manual (the second row) segmentations of three test samples in Group 1. The processing time of the automatic segmentation on one image was approximately one minute per CBCT image set, using a GPU with a model of NVIDIA GTX 2080 Ti. Although the automatic results still needed an orthodontist to refine the images (average of approximately 5 minutes per CBCT image), the total processing time to obtain the final accurate segmentation results for Group 2 was significantly reduced, compared to the time for Group 1 (average of approximately 10 hours per CBCT images).

Figure 2. — The segmentation results of the three test samples in Group 1. The first and second rows represent the results from 3D U-Net and manual, respectively. The red and orange parts represent the maxillae, and the green and yellow parts represent the cleft defects.

Maxillary Asymmetry Analysis and Defect Measurements

Statistically significant differences were observed upon maxillary asymmetry analysis, shown in Table 2. An overall difference in the measurements was noted on the cleft side of the maxilla, mostly concerning the pyriform aperture and alveolar crest area. The cleft side of the maxilla demonstrated significantly reduced values of maxillary volume (V_max) and length (L_max) as well as alveolar length (L_alv), anterior width (AntW_alv), posterior width (PosW_alv), anterior height (AntH_alv), and posterior height (PosH_alv). A significant increase in maxillary anterior width (AntW_max) was observed for the cleft side of the maxilla. The defect volume (V_def), length (L_def), width (W_def), and height (H_def) had mean measurements of 1.24±0.29×10³mm³, 18.76±7.25mm, 13.89±1.89mm, 13.40±3.10mm, respectively. The defect structure parameters are shown in Table 3.

Table 2:

Measurement and analysis of the cleft and non-cleft sides of the maxilla.

Parameter	Defect side		Non-defect side		P-value
Parameter	Mean	SD	Mean	SD	P-value
V_max (×10³ mm³)	18.02	3.24	19.32	3.53	0.000^*
L_max (mm)	34.31	2.56	41.27	3.72	0.000^*
AntW_max (mm)	14.42	2.55	13.09	1.95	0.002^*
PosW_max (mm)	34.73	2.91	35.08	2.44	0.293
AntH_max (mm)	20.68	3.98	21.33	4.67	0.236
PosH_max (mm)	32.91	4.31	32.89	4.33	0.893
L_alv (mm)	36.81	3.95	42.27	5.25	0.000^*
AntW_alv (mm)	15.91	2.21	17.87	1.99	0.000^*
PosW_alv (mm)	27.80	2.10	28.79	1.83	0.000^*
AntH_alv (mm)	14.45	3.15	15.45	2.76	0.000^*
PosH_alv (mm)	10.39	3.19	11.13	3.38	0.009^*

Open in a new tab

Significant at P-value<0.05

Table 3:

Defect structure parameters and measurements, where V_def denotes the volume of the segmented defect; and L_def, W_def, and H_def denote the maximum sagittal, transverse, and vertical distances of the defect, respectively.

Parameter	Mean	SD
V_def (×10³ mm³)	1.24	0.29
L_def(mm)	18.76	7.25
W_def(mm)	13.89	1.89
H_def(mm)	13.40	3.10

Open in a new tab

Multiple Linear Regression

Since cleft lip and palate can be caused by genetics, we considered defect parameters as independent variables and maxillary parameters on the cleft side as dependent variables. Each analysis was adjusted for age and gender. After performing a multiple linear regression followed by the Bonferroni correction, it was found that (1) the cleft side maxillary volume (V_max), maxillary length (L_max), alveolar anterior height (AntH_alv), and alveolar posterior height (PosH_alv) were significantly related to the defect height (H_def); (2) the maxillary anterior width (AntW_max), alveolar anterior width (AntW_alv), and alveolar posterior width (PosW_alv) were significantly related to the defect width (W_def); (3) the alveolar length (L_alv) was significantly related to the defect height (H_def); and (4) the alveolar anterior width (AntW_alv) was also significantly related to the defect volume (V_def). The linear regression analysis results between the maxillary and defect parameters on the cleft side are presented in Table 4.

Table 4:

Results of multiple linear regression analysis regarding the defect and relationship to the cleft side of the maxilla with adjusted age and gender.

Dependent variable	Independent variable	Coefficient	Standard error	P-value	R² adjusted
V_max (×10³ mm³)	V_def (×10³ mm³)	1.440	1.240	0.252	0.474
	L_def(mm)	−0.067	0.044	0.135
	W_def(mm)	0.307	0.190	0.113
	H_def(mm)	0.329	0.108	0.003^*
L_max (mm)	V_def (×10³ mm³)	−1.304	1.156	0.264	0.272
	L_def(mm)	−0.029	0.041	0.485
	W_def(mm)	0.273	0.177	0.128
	H_def(mm)	0.314	0.100	0.003^*
AntW_max (mm)	V_def (×10³ mm³)	−1.487	1.224	0.230	0.180
	L_def(mm)	−0.037	0.043	0.394
	W_def(mm)	0.584	0.187	0.003^*
	H_def(mm)	−0.034	0.106	0.747
PosW_max (mm)	V_def (×10³ mm³)	−1.652	1.375	0.235	0.201
	L_def(mm)	−0.015	0.049	0.757
	W_def(mm)	0.553	0.210	0.011^†
	H_def(mm)	0.214	0.199	0.077
AntH_max (mm)	V_def (×10³ mm³)	1.263	1.556	0.421	0.512
	L_def(mm)	−0.018	0.055	0.749
	W_def(mm)	0.322	0.238	0.182
	H_def(mm)	−0.096	0.134	0.480
PosH_max (mm)	V_def (×10³ mm³)	3.059	1.957	0.124	0.265
	L_def(mm)	−0.091	0.069	0.194
	W_def(mm)	−0.167	0.299	0.580
	H_def(mm)	0.124	0.169	0.466
L_alv (mm)	V_def (×10³ mm³)	−1.286	1.673	0.446	0.359
	L_def(mm)	−0.132	0.059	0.031^†
	W_def(mm)	0.526	0.256	0.045^†
	H_def(mm)	0.427	0.145	0.005^*
AntW_alv (mm)	V_def (×10³ mm³)	−2.916	0.961	0.004^*	0.325
	L_def(mm)	−0.044	0.034	0.200
	W_def(mm)	0.615	0.147	0.000^*
	H_def(mm)	−0.053	0.083	0.524
PosW_alv (mm)	V_def (×10³ mm³)	0.446	1.016	0.662	0.164
	L_def(mm)	−0.023	0.036	0.520
	W_def(mm)	0.477	0.155	0.003^*
	H_def(mm)	0.111	0.088	0.213
AntH_alv (mm)	V_def (×10³ mm³)	0.828	1.405	0.558	0.293
	L_def(mm)	0.009	0.05	0.851
	W_def(mm)	0.082	0.215	0.705
	H_def(mm)	0.596	0.121	0.000^*
PosH_alv (mm)	V_def (×10³ mm³)	−1.466	1.313	0.269	0.396
	L_def(mm)	0.004	0.047	0.924
	W_def(mm)	0.339	0.201	0.097
	H_def(mm)	0.331	0.114	0.005^*

Open in a new tab

Significant at P-value<0.05

^†

Significant at P-value<0.05 but >0.05 after adjustment by the Bonferroni correction.

DISCUSSION

This study conducted auto-segmentation of the maxilla and defect using CBCT images, quantified the 3D asymmetry of the maxilla in patients with UCP, and investigated the defect factors responsible for the variability of the maxilla on the cleft side. Achieving a deeper understanding of maxilla variability and the cleft defect’s anatomical relationship with the maxilla was the major goal, contributing to the limited body of information regarding the 3D assessment of craniofacial anomalies. This study had a relatively large sample size, with 60 CBCT image sets. Although many genetic and environmental factors related to cleft lip and palate have been identified²¹, there are still many unknown underlying mechanisms.

In order to quantify the maxilla and cleft in three dimensions, accurate and efficient CBCT image segmentation is the most fundamental task. Although manual segmentation is considered as the gold standard, this process requires several hours depending on the structure and range of region of interest (ROI). This drawback limits its clinical application. Recently, deep learning-based methods have been developed for automated volumetric segmentation. Minnema et al. compared three different convolutional neural networks (CNNs) for the lower part of maxilla and mandible segmentations and demonstrated DSCs around 0.87.²² For cleft volume estimation, Zhang et al. proposed a non-rigid registration technique, which explores the discrepancy between the maxilla from cleft lip and palate (CLP) patients’ CBCTs and the complete maxilla from template CBCTs, working with the 3D U-Net to achieve DSCs of 0.88 and 0.83 for the maxilla and cleft, based on 21 CBCT images of unilateral and bilateral CLP patients. Clearly, their advanced method outperforms our automated segmentation results, which were from pure 3D U-Net.¹⁶ However, this state-of-the-art deep learning algorithm’s accuracy is still not satisfied with the clinical retrospective study. In contrast, we jointly used manual and automated methods. We used the deep learning method to obtain initial segmentations and then manually refined them to accurately quantify the maxilla and cleft for the statistical analysis.

With the aid of CBCT, it is generally recognized that the cleft side of the maxilla is obviously hypoplastic, although it has not been fully analyzed in previous literature. In this study, the quantitative assessment indicated that the maxillary volume, length, and anterior width, along with all of the parameters of the alveolar crest, were significant different between the cleft and non-cleft sides. Furthermore, the maxillary anterior height, posterior width, and height demonstrated no statistically significant differences on either side. The unilateral pyriform aperture and alveolar crest collapse is a common morphological feature in UCP patients. This phenomenon could be due to the presence of a defect in the maxilla, which is adjacent to the aforementioned structures. There is no evidence that the maxillary structures far from the defect are affected. Suri et al.²³ and Li et al.²⁴ compared the asymmetry of the cleft and non-cleft sides using axial CT and spiral CT, respectively. They both indicated that most asymmetries and deformities were in the dentoalveolar area near the cleft and nasal chamber and not in the deeper regions of the maxillary complex. In agreement with the results of Agarwal et al.²⁵, this study found that the reduced maxillary volume on the cleft side was accurately reflected in the hypoplasia of the affected maxilla. Similar results were also described in our preliminary study.²⁶ The asymmetry of the pyriform aperture and alveolar crest is highly responsible for the aesthetics of the face. Thus, it is necessary to pay attention to this particular area while planning treatment for UCP patients, such as alveolar bone grafts, palatal expansion, and maxillary protraction. With regard to the defect, it is difficult to quantitatively analyze its dimensions due to its irregular shape and lack of anatomical landmarks. Previous studies of UCP patients mainly focus on the defect volume. Choi et al.²⁷ described an average defect volume of 1.3±0.5 ×10³ mm³ in UCP patients with a mean age of 9.8 years, which mirrors our results.

As stated in previous studies,^8,9 determining which factors are responsible for maxillary variability has proved challenging. However, the presence of a cleft is considered the initiating factor of maxillary dysplasia. In this study, a multiple linear regression was carried out to investigate the potential relationship between the parameters of the defect and those of the cleft side of the maxilla. With the adjusted age and gender of the subjects, the results indicated that there was a close relationship between them (see Table 4). In general, corresponding to the results of maxillary asymmetry analysis, there is a growing trend demonstrating that the larger the defect size, the larger the maxillary size. Among the results, the defect height was considered to be a crucial factor in the variability of the maxillary volume and length. The maxillary anterior width and posterior width exhibited a correlation with the defect width. As for the alveolar crest region, with the variation of parameters of the defect in the horizontal, midsagittal, and coronal planes, those of the alveolar crest were also changed accordingly. These could be explained by the previous growth of the maxilla. The cleft becomes larger with the growth of the maxilla, although the presence of the defect itself limits the overall growth of the maxilla. However, it should be noted that the low adjusted R² values suggest a relatively weak relationship between the maxilla and defect, as shown in Table 4. Thus, more variables should be taken into account in future studies to identify factors responsible for the variability of the maxilla.

CONCLUSIONS

The deep learning-based protocol could automatically segment the maxilla and defect with accuracy and efficiency and shows potential for large-scale clinical application in the future.
Significant hypoplasia of the maxilla on the cleft side existed mainly in the pyriform aperture and alveolar crest area near the defect.
The defect structures appeared to contribute to the variability of the maxilla, though additional studies are needed.

Acknowledgement

This work is supported, in part, by Ohio State University College of Dentistry, NIH/NIDCR DE022816, and NSF#1938533. We also gratefully acknowledge the support of computing resource provided by the Ohio Supercomputer Center.

Footnotes

Conflict of interest

The authors state that they had no conflict of interest during the conduct of the study.

DATA AVAILABLIITY

The data are not publicly available due to privacy and ethical restrictions.

REFERENCES

1.Prevalence at Birth of Cleft Lip with or without Cleft Palate: Data from the International Perinatal Database of Typical Oral Clefts (IPDTOC). Cleft Palate-Craniofacial J 2011;48(1):66–81. [DOI] [PubMed] [Google Scholar]
2.Hennekam RCM, Allanson JE, Krantz ID. Gorlin’s Syndromes of the Head and Neck Oxford University Press; 2010. [Google Scholar]
3.Taib BG, Taib AG, Swift AC, van Eeden S. Cleft lip and palate: diagnosis and management. Br J Hosp Med 2015;76(10):584–591. [DOI] [PubMed] [Google Scholar]
4.Patel DS, Jacobson R, Duan Y, Zhao L, Morris D, Cohen MN. Cleft Skeletal Asymmetry: Asymmetry Index, Classification and Application. Cleft Palate-Craniofacial J 2017;55(3):348–355. [DOI] [PubMed] [Google Scholar]
5.Yang L, Chen Z, Zhang X. A cone-beam computed tomography evaluation of facial asymmetry in unilateral cleft lip and palate individuals. J Oral Sci 2016;58(1):109–115. [DOI] [PubMed] [Google Scholar]
6.Choi Y-K, Park S-B, Kim Y-I, Son W-S. Three-dimensional evaluation of midfacial asymmetry in patients with nonsyndromic unilateral cleft lip and palate by cone-beam computed tomography. Korean J Orthod 2013;43(3):113–119. [DOI] [PMC free article] [PubMed] [Google Scholar]
7.Yoon Y-J, Perkiomaki MR, Tallents RH, et al. Association of Nasomaxillary Asymmetry in Children with Unilateral Cleft Lip and Palate and Their Parents. Cleft Palate-Craniofacial J 2003;40(5):493–497. [DOI] [PubMed] [Google Scholar]
8.Hall BK, Precious DS. Cleft lip, nose, and palate: the nasal septum as the pacemaker for midfacial growth. Oral Surg Oral Med Oral Pathol Oral Radiol 2013;115(4):442–447. [DOI] [PubMed] [Google Scholar]
9.Shi B, Losee JE. The impact of cleft lip and palate repair on maxillofacial growth. Int J Oral Sci 2015;7(1):14–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
10.Barbosa GL deR, Emodi O, Pretti H, et al. GAND classification and volumetric assessment of unilateral cleft lip and palate malformations using cone beam computed tomography. Int J Oral Maxillofac Surg 2016;45(11):1333–1340. [DOI] [PubMed] [Google Scholar]
11.Abdolali F, Zoroofi RA, Otake Y, Sato Y. A novel image-based retrieval system for characterization of maxillofacial lesions in cone beam CT images. Int J Comput Assist Radiol Surg. 2019;14(5):785–796. [DOI] [PubMed] [Google Scholar]
12.Hwang J-J, Jung Y-H, Cho B-H, Heo M-S. An overview of deep learning in the field of dentistry. Imaging Sci Dent 2019;49(1):1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]
13.Chang Y-B, Xia JJ, Yuan P, et al. 3D segmentation of maxilla in cone-beam computed tomography imaging using base invariant wavelet active shape model on customized two-manifold topology. J Xray Sci Technol 2013;21:251–282. [DOI] [PMC free article] [PubMed] [Google Scholar]
14.Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation BT - Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 In: Navab N, Hornegger J, Wells WM, Frangi AF, eds. Cham: Springer International Publishing; 2015:234–241. [Google Scholar]
15.Çiçek Ö, Abdulkadir A, Lienkamp SS, Brox T, Ronneberger O. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation BT - Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016 In: Ourselin S, Joskowicz L, Sabuncu MR, Unal G, Wells W, eds. Cham: Springer International Publishing; 2016:424–432. [Google Scholar]
16.Zhang Y, Pei Y, Chen S, et al. Volumetric Registration-Based Cleft Volume Estimation of Alveolar Cleft Grafting Procedures. In: 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI).; 2020:99–103. [Google Scholar]
17.Yushkevich PA, Piven J, Hazlett HC, et al. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. Neuroimage 2006;31(3):1116–1128. [DOI] [PubMed] [Google Scholar]
18.Sudre CH, Li W, Vercauteren T, Ourselin S, Jorge Cardoso M. Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations BT - Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support In: Cardoso MJ, Arbel T, Carneiro G, et al. , eds. Cham: Springer International Publishing; 2017:240–248. [DOI] [PMC free article] [PubMed] [Google Scholar]
19.Reddi SJ, Kale S, Kumar S. On the Convergence of Adam and Beyond. eprint arXiv:190409237 April 2019:arXiv:1904.09237.
20.Paszke A, Gross S, Massa F, et al. Pytorch: An imperative style, high-performance deep learning library. In: Advances in Neural Information Processing Systems.; 2019:8026–8037. [Google Scholar]
21.Leslie EJ, Marazita ML. Genetics of Cleft Lip and Cleft Palate. Am J Med Genet C Semin Med Genet 2013;163(4):246–258. [DOI] [PMC free article] [PubMed] [Google Scholar]
22.Minnema J, van Eijnatten M, Hendriksen AA, et al. Segmentation of dental cone-beam CT scans affected by metal artifacts using a mixed-scale dense convolutional neural network. Med Phys 2019;46(11):5027–5035. [DOI] [PMC free article] [PubMed] [Google Scholar]
23.Suri S, Utreja A, Khandelwal N, Mago SK. Craniofacial Computerized Tomography Analysis of the midface of patients with repaired complete unilateral cleft lip and palate. Am J Orthod Dentofac Orthop 2008;134(3):418–429. [DOI] [PubMed] [Google Scholar]
24.Li H, Yang Y, Chen Y, et al. Three-Dimensional Reconstruction of Maxillae Using Spiral Computed Tomography and Its Application in Postoperative Adult Patients With Unilateral Complete Cleft Lip and Palate. J Oral Maxillofac Surg 2011;69(12):e549–e557. [DOI] [PubMed] [Google Scholar]
25.Agarwal R, Parihar A, Mandhani PA, Chandra R. Three-Dimensional Computed Tomographic Analysis of the Maxilla in Unilateral Cleft Lip and Palate: Implications for Rhinoplasty. J Craniofac Surg 2012;23(5). [DOI] [PubMed] [Google Scholar]
26.Wang X, Zhang M, Han J, Wang H, Li S. Three-dimensional evaluation of maxillary sinus and maxilla for adolescent patients with unilateral cleft lip and palate using cone-beam computed tomography. Int J Pediatr Otorhinolaryngol 2020;135:110085. [DOI] [PubMed] [Google Scholar]
27.Choi HS, Choi HG, Kim SH, et al. Influence of the Alveolar Cleft Type on Preoperative Estimation Using 3D CT Assessment for Alveolar Cleft. Arch Plast Surg 2012;39(5):477–482. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R1] 1.Prevalence at Birth of Cleft Lip with or without Cleft Palate: Data from the International Perinatal Database of Typical Oral Clefts (IPDTOC). Cleft Palate-Craniofacial J 2011;48(1):66–81. [DOI] [PubMed] [Google Scholar]

[R2] 2.Hennekam RCM, Allanson JE, Krantz ID. Gorlin’s Syndromes of the Head and Neck Oxford University Press; 2010. [Google Scholar]

[R3] 3.Taib BG, Taib AG, Swift AC, van Eeden S. Cleft lip and palate: diagnosis and management. Br J Hosp Med 2015;76(10):584–591. [DOI] [PubMed] [Google Scholar]

[R4] 4.Patel DS, Jacobson R, Duan Y, Zhao L, Morris D, Cohen MN. Cleft Skeletal Asymmetry: Asymmetry Index, Classification and Application. Cleft Palate-Craniofacial J 2017;55(3):348–355. [DOI] [PubMed] [Google Scholar]

[R5] 5.Yang L, Chen Z, Zhang X. A cone-beam computed tomography evaluation of facial asymmetry in unilateral cleft lip and palate individuals. J Oral Sci 2016;58(1):109–115. [DOI] [PubMed] [Google Scholar]

[R6] 6.Choi Y-K, Park S-B, Kim Y-I, Son W-S. Three-dimensional evaluation of midfacial asymmetry in patients with nonsyndromic unilateral cleft lip and palate by cone-beam computed tomography. Korean J Orthod 2013;43(3):113–119. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R7] 7.Yoon Y-J, Perkiomaki MR, Tallents RH, et al. Association of Nasomaxillary Asymmetry in Children with Unilateral Cleft Lip and Palate and Their Parents. Cleft Palate-Craniofacial J 2003;40(5):493–497. [DOI] [PubMed] [Google Scholar]

[R8] 8.Hall BK, Precious DS. Cleft lip, nose, and palate: the nasal septum as the pacemaker for midfacial growth. Oral Surg Oral Med Oral Pathol Oral Radiol 2013;115(4):442–447. [DOI] [PubMed] [Google Scholar]

[R9] 9.Shi B, Losee JE. The impact of cleft lip and palate repair on maxillofacial growth. Int J Oral Sci 2015;7(1):14–17. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R10] 10.Barbosa GL deR, Emodi O, Pretti H, et al. GAND classification and volumetric assessment of unilateral cleft lip and palate malformations using cone beam computed tomography. Int J Oral Maxillofac Surg 2016;45(11):1333–1340. [DOI] [PubMed] [Google Scholar]

[R11] 11.Abdolali F, Zoroofi RA, Otake Y, Sato Y. A novel image-based retrieval system for characterization of maxillofacial lesions in cone beam CT images. Int J Comput Assist Radiol Surg. 2019;14(5):785–796. [DOI] [PubMed] [Google Scholar]

[R12] 12.Hwang J-J, Jung Y-H, Cho B-H, Heo M-S. An overview of deep learning in the field of dentistry. Imaging Sci Dent 2019;49(1):1–7. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R13] 13.Chang Y-B, Xia JJ, Yuan P, et al. 3D segmentation of maxilla in cone-beam computed tomography imaging using base invariant wavelet active shape model on customized two-manifold topology. J Xray Sci Technol 2013;21:251–282. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R14] 14.Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation BT - Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 In: Navab N, Hornegger J, Wells WM, Frangi AF, eds. Cham: Springer International Publishing; 2015:234–241. [Google Scholar]

[R15] 15.Çiçek Ö, Abdulkadir A, Lienkamp SS, Brox T, Ronneberger O. 3D U-Net: Learning Dense Volumetric Segmentation from Sparse Annotation BT - Medical Image Computing and Computer-Assisted Intervention – MICCAI 2016 In: Ourselin S, Joskowicz L, Sabuncu MR, Unal G, Wells W, eds. Cham: Springer International Publishing; 2016:424–432. [Google Scholar]

[R16] 16.Zhang Y, Pei Y, Chen S, et al. Volumetric Registration-Based Cleft Volume Estimation of Alveolar Cleft Grafting Procedures. In: 2020 IEEE 17th International Symposium on Biomedical Imaging (ISBI).; 2020:99–103. [Google Scholar]

[R17] 17.Yushkevich PA, Piven J, Hazlett HC, et al. User-guided 3D active contour segmentation of anatomical structures: significantly improved efficiency and reliability. Neuroimage 2006;31(3):1116–1128. [DOI] [PubMed] [Google Scholar]

[R18] 18.Sudre CH, Li W, Vercauteren T, Ourselin S, Jorge Cardoso M. Generalised Dice Overlap as a Deep Learning Loss Function for Highly Unbalanced Segmentations BT - Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support In: Cardoso MJ, Arbel T, Carneiro G, et al. , eds. Cham: Springer International Publishing; 2017:240–248. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R19] 19.Reddi SJ, Kale S, Kumar S. On the Convergence of Adam and Beyond. eprint arXiv:190409237 April 2019:arXiv:1904.09237.

[R20] 20.Paszke A, Gross S, Massa F, et al. Pytorch: An imperative style, high-performance deep learning library. In: Advances in Neural Information Processing Systems.; 2019:8026–8037. [Google Scholar]

[R21] 21.Leslie EJ, Marazita ML. Genetics of Cleft Lip and Cleft Palate. Am J Med Genet C Semin Med Genet 2013;163(4):246–258. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R22] 22.Minnema J, van Eijnatten M, Hendriksen AA, et al. Segmentation of dental cone-beam CT scans affected by metal artifacts using a mixed-scale dense convolutional neural network. Med Phys 2019;46(11):5027–5035. [DOI] [PMC free article] [PubMed] [Google Scholar]

[R23] 23.Suri S, Utreja A, Khandelwal N, Mago SK. Craniofacial Computerized Tomography Analysis of the midface of patients with repaired complete unilateral cleft lip and palate. Am J Orthod Dentofac Orthop 2008;134(3):418–429. [DOI] [PubMed] [Google Scholar]

[R24] 24.Li H, Yang Y, Chen Y, et al. Three-Dimensional Reconstruction of Maxillae Using Spiral Computed Tomography and Its Application in Postoperative Adult Patients With Unilateral Complete Cleft Lip and Palate. J Oral Maxillofac Surg 2011;69(12):e549–e557. [DOI] [PubMed] [Google Scholar]

[R25] 25.Agarwal R, Parihar A, Mandhani PA, Chandra R. Three-Dimensional Computed Tomographic Analysis of the Maxilla in Unilateral Cleft Lip and Palate: Implications for Rhinoplasty. J Craniofac Surg 2012;23(5). [DOI] [PubMed] [Google Scholar]

[R26] 26.Wang X, Zhang M, Han J, Wang H, Li S. Three-dimensional evaluation of maxillary sinus and maxilla for adolescent patients with unilateral cleft lip and palate using cone-beam computed tomography. Int J Pediatr Otorhinolaryngol 2020;135:110085. [DOI] [PubMed] [Google Scholar]

[R27] 27.Choi HS, Choi HG, Kim SH, et al. Influence of the Alveolar Cleft Type on Preoperative Estimation Using 3D CT Assessment for Alveolar Cleft. Arch Plast Surg 2012;39(5):477–482. [DOI] [PMC free article] [PubMed] [Google Scholar]

PERMALINK

3D morphometric quantification of maxillae and defects for patients with unilateral cleft palate via deep learning-based CBCT image auto-segmentation

Xiaoyu Wang

Matthew Pastewait

Tai-Hsien Wu

Chunfeng Lian

Beatriz Tejera

Yan-Ting Lee

Feng-Chang Lin

Li Wang

Dinggang Shen

Song Li

Ching-Chang Ko

Abstract

Objective:

Setting and Sample Population:

Materials and Methods:

Results:

Conclusions:

INTRODUCTION

MATERIALS AND METHODS

Subjects

Maxilla and Defect Segmentation

Description of Measurement

Table 1:

Figure 1:

Statistical Analysis

RESULTS

Maxilla and Defect Segmentation

Figure 2.

Maxillary Asymmetry Analysis and Defect Measurements

Table 2:

Table 3:

Multiple Linear Regression

Table 4:

DISCUSSION

CONCLUSIONS

Acknowledgement

Footnotes

REFERENCES

ACTIONS

PERMALINK

RESOURCES

Similar articles

Cited by other articles

Links to NCBI Databases