DOI: 10.1051/odfen/2018117 J Dentofacial Anom Orthod 2016;19:409 © The authors

3D cephalometry and artificial intelligence

J. Faure1, A. Oueiss2, J. Treil3, S. Chen4, V. Wong4, J.-M. Inglese4 1 DFO Specialist, University Lecturer and Hospital Practitioner, Private Practice 2 DFO Assistant, Nice. DDS, Dip. Orthodontics (Toulouse III, Anthropobiology) 3 Neuroradiologist (Pasteur Clinic, Toulouse) 4 Research and Development Department, Carestream Health Rochester NY 14608 USA

SUMMARY Orthodontists today work more and more in a three-dimensional (3D) universe with cone-beam exam- inations occurring more frequently, now supplemented by digital prints and 3D portraits. So far these documents are used primarily as esthetic imagery; superimposition techniques, issued from geometric morphometrics, allow a pseudoquantified approach. The implementation of true cephalic biometrics requires consideration of the complete craniofacial set at different anatomical levels (alveolodental/basic bone/frame or overall architecture) and in three dimensions. It must lead to a quantified description of the anatomy, dysmorphism, and the necessary therapy to correct it. A parametric approach is needed to choose the landmarking, the definition of the orthogonal refer- ence, the definition, and selection of parameters. Given the number of parameters required for a description without fault, the use of a simple tool with artificial intelligence is inevitable.

KEYWORDS 3D cephalometry, 3D biometrics, dental landmarks, bone markers, choice of parameters, artificial intelligence

INTRODUCTION

For many years now, orthodontists have The fundamental challenge was the ability been trying to achieve a three-dimensional to easily identify the anatomical landmarks (3D) approach to the craniofacial system. To on the three snapshots. Consequently, this end, in the past they would take three points A and B would be relatively discern- conventional cephalometric photographs ible on the snapshot in the lateral view but taken in three dimensions: the lateral inci- impossible to identify in the basal and fron- dence and profile cephalometry, the basal tal images while the transverse position incidence and basal cephalometry, and the remained impossible to pinpoint. frontal incidence and frontal cephalometry.

Address for correspondence: Jacques Faure 10, place Lannes – 32021 Auch – France Article received: 21-04-2016. E-mail: [email protected] Accepted for publication: 30-05-2016.

This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. 1

Article available at https://www.jdao-journal.org or https://doi.org/10.1051/odfen/2018117 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

Within recent times, the advent Luebbert et al.31 use the following of the scanner and cone-beam x-ray distances: has brought the orthodontist into the – Between the apices of the central 3D realm. Thereafter, other 3D tools incisors, those of the vestibular appeared; these capture the den- roots of the first premolars, and tal arches and cutaneous envelope those of the vestibulomesial roots with remarkable precision and color of the first maxillary molars. rendering. – Between the pulp chambers of From then on, the manipulation of 3D these same teeth. images pervaded our daily work and – Finally, between the maxillary exter- now 3D cephalometry is continuously nal cortical walls at the apex of the made available to us and can be used vestibular and vestibulomesial roots at the clinic. of both the premolars and the first Of course, all 3D input tools have molars. easy-to-use measuring instruments. For example, a digital impression scan- It is obvious that these measure- ner always uses a tool that measures ments have no daily clinical use, the distance between two points, you these parameters were chosen with a simply have to point at landmarks P research objective, without any diag- and Q and the software immediately nostic benefits. The “landmark” sites, displays the PQ measurement. Similar- particularly the external cortex of the ly, there are instruments for measuring alveolar bone close to the apex, have the angle formed by two intersecting neither the physiological identity nor lines, the angle formed by two planes, the anatomical specificity to permit a the cartesian coordinates, etc. diagnostic analysis. Most often, however, the use of 3D To date, a complete software that tools is limited to the identification of a can be used daily at clinic, which can particular anatomical element (an impact- cover the entire craniofacial system ed tooth for example or odontoma). in the three dimensions of space, Though rare, authors advocate using including the alveolar, dental, basal, the unique measurements or pseu- and architectural floors (or maxillo- do-measurements which result from facial framework), has not yet been the techniques of superimposition. made available. Such a program will, The difference between the individuals of course, have to precisely describe compared (for example the subject’s the anatomy and therefore be able to measurements before treatment and quantify the dysmorphia to be used after surgery) is expressed by a colori- as a guide in the establishment of the metric scale. treatment plan. Finally, we can find articles where In the following sections, we will effective measurements were achieved examine the current uses of 3D imag- using the software tools of the scan- ing for therapeutic purposes and ner or cone-beam x-ray. For example, discuss the constraints and require- in a study on the skeletal effects of ments for a complete and coherent 3D maxillary disjunction by Hyrax, authors cephalometry.

2 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

3D IMAGING AT THE CLINIC: THE USE OF GEOMETRIC MORPHOMETRIC TOOLS

The use of 3D imaging in standard or – The nature of the phenomenon stud- clinical research is currently giving way ied. If the evolution is global, a growth to the geometric morphometric tools process for example, the Procustes of superimposition, either by the direct superimposition will be the desired use of geometric morphological soft- tool. If, however, the evolution is well ware, (Morphological, Rapid Form etc.) localized, for example a mandibular or by integrating analog modules into advancement surgery, a superimpo- the orthodontic or surgical software. sition on the “recorded area” (in this case the ascending ramus posterior Using geometric morphometry: to the osteotomy), will be preferred. the Procustes superimposition – The isolable or “dissectable” nature of the area of interest (actual or vir- Firstly, these tools are used when tual dissection). comparing two individuals: A and B. – The existence and distribution In conventional superimpositions, of landmarks or superimposition the user identifies an area that is zones, on the stable site (recorded assumed to be a stable, “recorded area) and on the supposedly mobile area” and thus visualizes the evolu- site. tion (growth, therapeutic, maturation, etc.) of the areas that are modified in The boundary is not as rigid as it relation to this stable reference area. may seem. With the distribution of the It is therefore an operator-dependent inhomogeneous landmarks, Procustes choice. With the Procustes superim- superimposition will begin to favor the position, the software corrects the area of maximum landmark density, or scale, orienting the two subjects by recorded pseudo-zone. parallelizing the axes of inertia and Procustes superimposition has superimposing as precisely as possi- evolved conventional methods by facil- ble, the minimum distances between itating superimposition on a stable, the homologous points of subjects A recorded area which is often sought and B, on the entire subject studied. after and is considered a recalibration. This technique separates size effects In a maxillofacial surgical study, an from form effects and is non-operator area of superimposition is often taken dependent. in the anterior region of the base Superimposition with a supposedly and the tools used to assess the A/B stable recorded area, will better demon- deviations, standard Procustes super- strate the variation observed within imposition tools will demonstrate the range of this area rather than a gener- difference with the skull base recorded alized Procustes superimposition. and unchanged. The choice between these two Oueiss20,36,37 recalibrated points SO, methods will of course depend on the IO, HM (table I), to study the asym ­­ following: metry.

J Dentofacial Anom Orthod 2016;19:409 3 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

Table I: The 14 trigeminal points. The eight points of the Treil framework model of the skeletal framework or the global architecture: RHM, LHM, RSO, LSO, RIO, LIO, RM, LM. The landmarks of the maxillary or mandibular ­bases are represented by: RM, LM, RFM, LFM, UNP, LNP, RGP, LGP. (The mental points are both frame and maxillary base landmarks.)

Environment Skeletal Landmarks

RHMLHM MHM Hammer Head: right and le Frame

RSO LSO MSO Supraorbital: right and le Frame

RIO LIO MIO Infraorbital: right and le Frame

RM LM MM Chin: right and le Frame and base

RFMLFM MFM Mandibular foramen: right and le Base

UNP LNP MNP Nasopalataline foramen: upper and lower Base

RGPLGP MGPGreater palane canal: right and le Base

Landmarking: what to immediately identifiable anatomical superimpose on? detail (pseudo-landmarks or semi-land- marks, Bookstein5,6,7). The description of an object’s anato- Moreover, whatever the mode of ana- my is either represented by a diagram tomical identification, (point diagram or of characteristic points or landmarks, direct surface capture) it is possible to (each one assigned to its cartesian achieve a global superimposition over coordinates) or on the complete 3D the entire anatomical element studied, capture of the surface or a specific or to determine a stable area of super- portion of it. The A/B comparison will imposition “the recorded area.” involve the superimposition of two- What type of 3D capture? point diagrams or the direct superim- position of surfaces A and B. Dentoskeletal The choice of technique depends on the existence or absence of eas- For a predominantly voluminal record- ily identifiable anatomical landmark ing, the Rx method was previously points. In any case, the reproducibility used but now cone-beam CT scans are and accuracy of the landmarks must required. first be assessed48,35,30. In the absence of characteristic points, it is possible to Dental occlusal use meshing methods to create virtual To capture the surface of the arch, benchmarks distributed ­homogenously the intraoral scanner can be used on a common surface without any to scan the arch very quickly, with

4 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

remarkable color and texture precision. color for teeth and occlusions as well It also allows for a reliable recording of as for the face and the smile. the occlusal ratios. Together they provide a complete understanding of the evolution of the Mucocutaneous occlusion, cutaneous surface, and There are several possible methods skeletal architecture in addition to their for capturing the 3D surface of the ratios. A superimposition on the skull facial skin or mucosal coating: base can be used to assess the cuta- neous and occlusal changes in relation – Photogrammetry, derived from the to the orthodontic procedures, ther- principles of stereographic pho- apeutic tooth displacements, and/or tography, a technique widely used surgical displacements. in cartography10,11 It is possible to only consider the – The optical scanner10,11 cutaneous variations or even the iso- – Structured light 3D scanners lated dental occlusal variations when – Radiographic techniques such as superimposing on the cutaneous 26,27,40 scanners and cone-beam computed “shell” alone : the observer will tomography not be able therefore to assess the – Or even MRI imaging, 3D ultra- causal role of the tooth and skeletal sound, and holography. displacements in possible cutaneous improvement. Similarly, we can identify However, the methods that provide and study cutaneous, ethnic, and mor- a complete recording of a digitized 3D phological differences without having surface are preferable to ­photographic the appropriate tools to decipher the techniques. The file can, therefore, be dental or skeletal roles in causation. combined with the dentoskeletal file delivered by cone-beam ­computed Which stable reference area tomography, by a fusion method, or “recorded area” should be which will eliminate the need for opti- chosen? cal methods such as stereographic photography. If we choose superimposition on a stable reference area there are inher- Fusion ent challenges. The ideal solution to obtain complete patient files is to conduct a cone-beam Superimposition on the bone examination, a digital impression (via structures an optical scanner), and a 3D facial To study the maxillofacial evolution, photograph. many authors use the cranial base, or The 3D images from the last two more specifically its anterior part, as examinations are then integrated and their area of superimposition. Cedi- combined in the 3D maxillofacial recon- vanes has defined a superimposition stitution of the cone beam by “fusion” protocol. On two cranial base images, methods32. This will compensate for A and B, arranged in the same virtu- insufficient resolution and the lack of al space, homologous landmarks are

J Dentofacial Anom Orthod 2016;19:409 5 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

identified. The program then deter- benchmarks inclusive of the internal mines the displacement matrix super- canthi of the eyes, commissures and imposing A on B. A manual adjustment nose tip.24-28 According to Moeren- is necessary. The A/B differences can hout et al. there are 13: the sutures, be quantified in the orthonormal sys- wings, internal canthi D and G, pupil- tem defined by the benchmark dia- laries D and G, subnasal gnathion, gram. The limits of this method are zygions D and G and the right and consistent with a system based on the left supraorbital points.33 Hoefert et comparison of two individuals. We can al. have identified seven perinasal also perform a “recalibration” on the points.21 orbital zone36,20, characterized by points However, for these methods, the IO, SO, and HM (table I). differences are generally color-coded and they are especially effective for Superimposition on the mucosal A/B comparison. structures, the palatine rugae Recently, a complex method devel- 34 To analyze tooth displacements dur- oped by Nanda uses tools derived ing therapy, we have superimposed from neuroimaging: each subject’s the images on casts; the first set face benefits from a topological exam- of images are two-dimensional (2D) ination (spherical mapping). We can, occlusal photographic images13. Then therefore, record each facial shape 3D photographs are used (photogram- and construct “an average face.” The metry and stereographic photography), quantifiable diagnosis is the result before an optical scanner is finally used of the “average” patient/subject to obtain digital casts4,30,39. The accura- superimposition which results in a cy of the digital cast compared to the color-coded visualization. plaster mold is demonstrated.29,30 Dig- ital casts prove to be extremely accu- Methods of assessing the rate for archiving and studying dental localized A/B discrepancy displacements. 2,12 In the literature, most authors pre- These methods borrowed from fer superimposing on the anterior geometric morphometry are ways mucosal folds of the palate, the pala- of comparing two individuals with tine rugae, and choose it as a reference similar morphologies, or methods of area because it is a stable site3,22,23,2,1,38, comparing the same object at various but others maintain a more critical stages of its evolution (growth, mat- perspective.41,14. uration, etc.) or at various stages of therapy (before, after treatment, after Superimposition of the cutaneous contention, relapse…) “shells” These methods are aimed at meas- uring the difference which quantifies To study the evolution of the cuta- the pathology and the normalizing neous “shell” different authors either act of correction. They are essentially advocate using a global superimposi- intended to assess the significance tion or specific benchmarks. Kau and of the difference. They can be used Richmond have identified five such for diagnostic purposes but only

6 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

indirectly as this is not their primary all correspond well to their capabilities. purpose. Alternative methods include: Though generally, the user tends to be – Colorimetric assessment of the dif- more concerned with the significance ferences between surfaces A and B, of the change rather than the quantifi- perpendicular to the surfaces (color- cation of the measured difference. ized visualization of the individual The comparison of the same subject differences Rapid Form Plus Pack 2). at stages T1 and T2 will reveal the sub- – Generalized Procustes Analysis tlest changes. (GPA, Dryden, and Mardia, 199315) However, the same is not true for its – Thin-plate analysis (Bookstein7) diagnostic uses: – Goodall test (Simple 3D) – Subject A must be compared to – Wilk’s lambda test (Morphologika2,1) a normomorphic subject to deter- – Principal Component Analysis (Jolife mine whether there is an active 1986, Dunteman, 1989) pathology. – EDMA: Euclydian Distance Matrix – The difference must be quantified Analysis. as a measure of the extent of the dysmorphia. For more details on these methods, – The treatment is therefore condi- the reader can consult the related tioned by the superimposition. studies20,37. A real normomorphic subject will be Which subjects should be difficult to find and a virtual one will superimposed? always be deprived of a little human- ity! Virtual normomorphic subjects can be derived from the average mor- The tools of geometric morphome- phology of a sample of normal sub- try are designed to analyze slight dif- jects; which itself is not easy to find ferences between individuals of the as there may be such challenges as: same or similar types? molar and canine class I; 2-mm incis- Their use in therapeutic follow-up, al overlap or overhang, no missing reviewing the therapeutic result, stud- teeth, no dentomaxillary disharmony ying growth as well as the initiating (DMD)24,27,28. and organizing therapeutic treatment,

CONSTRAINTS AND REQUIREMENTS FOR A TRUE THREE-DIMENSIONAL BIOMETRIC ANALYSIS

After having seen the limitations of and at the three anatomical levels (alve- the tools of geometrical morphometry. olar, basal, and skeletal framework). What follows will consider the con- It should permit a quantified descrip- straints imposed on the development tion of the dysmorphia, thereby allowing of a true 3D biometric analysis. a quantification of the treatment plan. As stated above, it must ensure that It should of course focus on the inter- the entire craniofacial system is account- arch ratios, which is the key objective of ed for in the three dimensions of space, any orthodontic therapeutic intervention.

J Dentofacial Anom Orthod 2016;19:409 7 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

Table II: The five functional categories of parameters. MNP: middle of the anterior palatal canal, identified by its superior and inferior orifices (UNP and LNP). MHM, MFM, and MM: respective media of the hammerheads, HM, mandibular foramina, FM and chin, M, right and left. GIMY: y coordinate of the centers of inertia of the maxillary and mandibular incisors. RGP-LGP and RFM-LFM: distances between the greater palatine canals and mandibular foramina D and G. TQIM: average torque of maxillary and mandibular incisive blocks.

.

.

We will successively consider land- – The chosen anatomical detail must marking, parameter selection, and arti- be identifiable without fail by any ficial intelligence, all of which assist in operator. accessing these data and in establish- ing a diagnosis and treatment plan. The osseous details (foramen) as they relate to the trigeminal system, Landmarking and choosing frames are guides of maxillomandibular con- of reference. junctival growth and constitute a choice without any other alternative. Skeletal Landmarks The 3D analysis program firstly relies There are two main criteria for choos- on 14 points, which define the skeletal ing skeletal landmarks: frame and mandibular base. – The chosen anatomical detail must The eight points of the primary anal- play a major role in the growth ysis of Treil highlight the bordering of 42,43,44,46,47 and physiology of the area to be the skeletal framework (Treil ) identified. and eight other points ensure the

8 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

identification or location of the bas- maxillomandibular ratios are signed dis- es (Faure16-19 and Oueiss37). The chin tances (+ for class-III ratios). points are mixed, they belong simulta- The use of a direct, orthonormal tri- neously to the framework and to the hedron associated with the anterior mandibular base. (Table I). region of the skull region, is necessary to assess the maxillofacial morphology. Identifying teeth: the inertia matrix Alternatively, to locate the dental or The mathematical identification of a paradental elements (reconstitution or solid object in space, as well as with- fixed prosthesis), one must adopt the in the maxillomandibular assembly, is dentist’s language: mesiodistal, vestib- done by calculating the inertia matrix. ulolingual, and infraocclusion–supraoc- This calculation defines the position of clusion; all of these refer to a system the center of inertia (Gx, Gy, and Gz) of curvilinear coordinates conforming and its orientation by the data of the to the shape of the arch. main axes of inertia. The use of inertia matrices is an “error-free” way of locat- Critique of the benchmarks produced ing not only individual teeth but also by 2D cephalometry groups of teeth which were joined or The appeal of this new technique lies fused together during the orthodontic mainly in its ability to capture all the mechanical procedure (Treil J45). standard points of conventional ceph- alometry (A, B, pogonion, chin, etc.) The futility of pure alveolar landmarks and as such there is no “break” in the Odontologists are well aware that user’s routine. the alveolar bone is totally dependent The method is simple: on the teeth. In orthodontics, anter- – Identify a point on the 2D profile oposterior measurements are based obtained from 3D input, on alveolar landmarks (A and B for for example the right suborbital example: SNA, SNB, and ANB) show point. maximum correlations and quasi-iden- – Then try to locate it in the right paras- tities, with measurements based on agittal plane on the 2D face-frontal dental landmarks (centers of inertia of X-ray obtained from the 3D input. maxillary groups, GIM and mandibular incisors, Gim: GIMy, Gimy, and GIMy- In addition to the severe inaccuracy, Gimy); This makes any choice of alve- there is a major ethical problem here— olar landmarks superfluous, ­especially why must the patient be subjected to because the dental landmarks are 3D capture only to derive clinical data more precisely identifiable. that is of limited use, inaccurate, and 2D.

Choosing frames of reference The Parameters Choosing a frame of reference is firstly Innumerable possibilities a mathematical obligation, because the algebraic calculations must be based on The first legitimate parameters are quantified and signed variables. Thus, the coordinates of the 14 skeletal the parameters of the anteroposterior points (14 × 3 = 42 parameters).

J Dentofacial Anom Orthod 2016;19:409 9 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

There are also many other parameters. 1) Direction: anteroposterior, vertical, – There are (32) parameters for indi- transverse: 3c vidual teeth and in terms of the 2) Pertinent/relevant anatomical ele- groups: primarily 12 upper and low- ment: alveolar/basal/skeletal frame- er incisors, two upper and lower work or architecture: A, B, C: 3c molars, right and left, four upper 3) Location of the area of interest with- and lower arches, two upper and in the anatomical element studied or lower right and left hemiarches, four position of the landmarks: anterior/ of which can be located by three middle/posterior: 3c linear parameters and three angu- 4) Measurement technique: 5c lar parameters, which gives a grand ◊ Direct measurement or function of total of 264 parameters. several parameters (example: co- – All the skeletal bipoints are fac- ordinate of the center of inertia of 2 ing point C14 , with each one hav- the incisor group or the maxillary ing three linear and three angular arch…) parameters: 91 points. ◊ The ratio of the linear (distance) – The triangular skeletal faces are and angular measurements: the 3, at point C14 with three linear and first form requires a scale correc- three angular parameters each: tion. 364 faces. ◊ Distance or vector (scalar distance or – If we add the quadrangular faces (for signed value), scalar angle, or signed example the lower mandibular face: angle. RFM, LFM, RM, and LM), and if we ◊ Cartesian and curvilinear coordi- introduce the interface or interpoint nates. angles, we will quickly exceed one ◊ Direct or projected measurements thousand parameters! (DEM for angles).

The functional parametric • Function classification According to the parameter’s func- Given the number of parameters like- tion (position R, morphology, B/H ratio, ly to provide an effective description of B/H pathological differential, common the maxillofacial anatomy, parameter high/low pathology): 5c. “recruitment” must be considered dif- ferently for individual cases, depending • Category enumeration on the functional needs. To this end, a Non-asymmetry/asymmetry: 2c functional or categorical classification Measurement mode (§ 1, 2, and 3), is required before the parameters are without considering the different tech- chosen. nical possibilities of measurement (§ 4): • There are two functional catego- 3 × 3 × 3 = 27c ries: asymmetrical parameters and non-asymmetrical parameters. Functions: 5c • Technical categories related to the A flawless description of the anatomy mode of measurement—these or maxillofacial pathology would neces- modes include: sitate as many as 270 parameters!

10 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

That equates to 270 parameters if we anteroposterior/basal (basal bone)/ only keep one parameter per category anterior landmarks. instead of proposing eight or 10 in each MNPy is a posteroanterior coor- (§ 4). dinate of the middle of the anterior palatal canal (anterior maxillary basal Note 1: concerning the types of landmarking). parameters Mm is a posteroanterior coordinate Among the types of parameters avail- between the mental foramen (anterior able to us, many are already present in mandibular basal landmarking). the usual 2D analyses, for example: MNPy-MMy: class-II discrepancy – Position (R): is the position of an between the maxillary and mandibular anatomical element or landmark/ bases, anterior segment. skull base: SNA, SNB, 1/Frankfort – B/H ratio: Adobos; ANB = SNA-SNB; Selection – morphology, gonial angle. A discriminant analysis or simple observation of the mean values of On the other hand, two types of the parameters studied in the ref- parameters are very effective and yet erence sample and in a pathological are only intuitively used without sys- sample (t-test) makes it possible to tematic “encryption.” identify the most effective parame- ters in the detection of major pathol- – Common pathology high-low: ogies. Stringent selection should (TqIM+Tqim)/2, angular biproalveolar. decrease the number of descriptive figures to approximately 100 while B/H pathology differential: (TqIM- ensuring a precise description of the Tqim)/class-II incisal torque differential. pathology. It creates or accentuates the occlusal For example, there are many meas- class II at the incisor level (or mod- urements of the same type. Thus, from erates class III in a class-III context: the posteroanterior position of the class-III compensation). A negative val- anterior basal mandibular region, the ue corresponds to a class-III differential possibilities are as follows: (or class-II compensation). MMy coordinate (R), , , , quadratic angles of the – Gimx is a position asymmetry param- faces RSO-LSO-RIO-LIO/RIO-LIO- eter of an anatomical/transverse/­ RM-LM, MHM-MM, MHM-MM/ alveolar/anterior landmark element MHM-MIO, MHM-MM all projected on Oy. The Gimx transverse coordinate (/R) Only one is kept: MMy is the most of the center of inertia of the lower statistically effective parameter for incisor group. differentiating a class-II or class-III MNPy-MM is a non-asymmetry dis- pathological group from the « normal » crepancy-based parameter: high/low, ­reference group.

J Dentofacial Anom Orthod 2016;19:409 11 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

Artificial Intelligence For example, a general schematic of Despite the statistical sorting and the class-III trend shows that the meas- the major elimination of many possi- ured clinical situation is characterized by: ble parameters, functional analysis has – a decreased overjet (underjet observed shown that it is impossible to decrease and measured); - molar ratios: (0) nor- the number of descriptive figures mal or neutral (measured class-I rela- (below 100) to 200. tion). The comment displayed must be This is hardly surprising because a “underjet despite class-I molar ratios.” conventional 2D analysis already con- tains about 20 measurements, includ- Clearly, if with two neighboring ing the anteroposterior orientation, the parameters there are nine possible incisor alveolar level, and one or two comments, then with three neighbor- vertical measurements. However, it ing parameters there will be 27 possi- does not account for basimaxillary or ble comments, etc. architectural landmarks; the transverse Artificial intelligence must be able to direction; or even transverse, vertical, provide a final synthesis—e.g., “excess or anteroposterior asymmetry. of the right hemiface” when there is Artificial intelligence is needed to a simultaneous right posteroanterior synthesize and produce a “written” excess and a right vertical excess on diagnosis. the left lateral . This program must first approximate To recognize a major syndrome, for the neighboring parameters: molar dis- example, this “excess of the right hemi- crepancy parameter (positive count for face” defined above, the software must class II), and B/H incisal discrepancy pair the skeletal parameters (of the frame parameter (positive count in class II), or bone base) in the three dimensions of and provide a comment. space.

CLINICAL CASE COMPARING THE IMPLEMENTATION OF GEOMETRIC MORPHOMETRIC METHODS AND A RIGOROUS PARAMETRIC CEPHALOMETRY

This patient came to our office having of the right hemiface. The diagno- transferred from a nearby town where sis includes a small alveolar class III, maxillary bands were placed without any a rotation of the mandibular arch to mention as to the asymmetry diagnosis. the left/maxillary (deviation from the centers and a more pronounced class Diagnosis III on the right), and a displacement to the left of the mandibular arch (slight Synthesized diagnosis: class-III hyper- right maxillary exoalveolar) (Figs. 1–4). divergence with excess asymmetry

12 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

Figure 1a, b, c Preliminary files. Portraits, profile, smiling face, face at rest (6/09; 15a 6m).

Figure 2 Figure 3 Portrait of the face at rest: facial lines Portrait of the face at rest: facial lines (6/09; 15a 6m). (6/09; 15a 6m).

Figure 4 preliminary files. Right, frontal, and left intraoral views. (6/09; 15a 6m).

J Dentofacial Anom Orthod 2016;19:409 13 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

There is a class-III skeletal pattern In addition, there is a “rolling” of the with a clear facial concavity; hyperdi- two arches, which is more significant vergence is responsible for an hourly in the lower arch, and affects the man- shift in the arches. dibular base. It is lower on the right Facial asymmetry, caused by a right compared to the left (vertical compen- hemiface excess, is characterized by sations are made by increasing the a maxillary deviation to the right and maxillary and mandibular alveolar pro- mandibular deviation to the left. cesses to the right). At the alveolar level, the left mandible Lastly, maxillary and mandibular experiences a characteristic deviation to anterior and lateral class-III compen- the left of the mandibular arch and a devi- sations and additional left lateral-­ ation to the left of the mandibular inter- mandibular compensations are noted incisal median (rotation to the left of the (Figs. 5, 6). mandibular arch in the horizontal plane.)

Figure 5 craniofacial scan number. Facial, profile, and basal views; with trigeminal benchmarks and the most significant asymmetry parameters.

14 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

Figure 6 The 8 trigeminal points diagram of Treil showing the general maxillo facial fram. Here a case of asymmetry and class III.

Treatment compensations (right posterior lower- ing of the maxilla along the right unilat- Treatment first involved a difficult eral vertical excess of the mandible has removal of anteroposterior compen- created a right posterior lateral infraoc- sations and asymmetry compen- clusion; Figs. 7–9). sations. The removal of the vertical

Figure 7a, b, c Right, frontal, and left intraoral views: removal of compensations. (11/09; 15a 11m).

J Dentofacial Anom Orthod 2016;19:409 15 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

Surgery included maxillary advance- ment, a more significant mandibular advancement on the left with a man- dibular inclination (elevation to the right of the mandible), and a genioplasty to reposition the chin (Figs. 10–13). Analysis of the dysmorphia

Figure 8 An analysis of the dysmorphia was Intraoral view: vertical decompensation. performed in 3D using the C2000 Cepha (02/10; 16a 2m). program, a parametric cephalometry

Figure 9a, b, c Preoperative craniofacial scan no. 2; frontal v4iew of face and the right and left profiles. (06/10; 16a 6m).

16 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

Figure 10a, b, c Pictures of the profile, smiling face, and the face at rest. Band removal. (12/10; age 17 years).

Figure 11a, b, c Right, frontal, and left intraoral views. Bands removed. (12/10; age 17 years).

Figure 12a, b, c Postoperative craniofacial scan no. 3. Frontal view of the face and the right and left profile views. Bands removed. (12/10; age 17 years).

J Dentofacial Anom Orthod 2016;19:409 17 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

Figure 13 Comparison of frontal views: early multiband treatment: preoperative and postoperative.

program that we used at that time. The treatment, (2) “follow” the overall 14-point trigeminal pattern and dental evolution or more specifically the final landmarking, determined by the inertia stage of the surgical evolution; color matrix calculation, were already per- coding the variations facilitates our formed in the parametric analysis. understanding. The use of geometric morphometric On the other hand, using these tools tools for superimposing on the orbital for diagnosis is virtually impossible in area (files 3/1, Rapid form, Fig. 14 or this instance. files 3/2, D1 mesh, Fig. 15) makes it The ar eas of significant orthodontic difficult to (1) comprehend the com- or surgical displacement (red T3/blue plexity of the pathology and its surgical T1, Fig. 14 Rapid form) can be seen.

Figure 14a, b, c, d Superimposition readjusted on the orbital sector (Rapid form): 1 = end, 3 = beginning. Right, frontal, left, and three-quarter views.

18 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

Figure 15a, b, c Recalculation of superimposed reconstructions: right, front, and left. End of treatment n. 3 /end of treatment n. 2 (Dl mesh).

− Incisor modifications: removal of − Chin modifications: repositioning compensations (high recoil, low ad- genioplasty vancement) and a surgical advance- − Gonial modifications: the left man- ment dibular advancement slightly “push- es out” the right gonial angle.

CONCLUSION

A true 3D biometric analysis needs to – Choices made are based on the sta- be developed. As we have discussed tistical analysis of the orthodontists’ above, there are rigorous and often specific hierarchy. contradictory constraints involved: – A selection of parameters allow- – The accurate identification of 3D ing an accurate description of any landmarks based on a stable ana- dysmorphia. tomical structure. An overview of the superimposition – Choosing numerous parameters to techniques most often used in a “pseu- ensure a complete description. dobiometric” analysis, shows the inad- – The fact that no more than 12 equacy of these tools and therefore parameters are necessary to give the need for a parametric approach. the reader the ability to easily dis- However, the latter in turn requires the cern at a single glance. use of artificial intelligence which has – Significant parameters are easily the following capabilities: comprehensible. – Synthesize and compare numerical – Linear measurement parameters values. are preferred and should be parallel – Manage the large number of to the orthonormal axes. parameters.

J Dentofacial Anom Orthod 2016;19:409 19 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

– Describe the dysmorphia in English Conflict of interest: JF, AO, and JT declare that not in figures. they have no conflict of interest. SC, VW, and – Only display the figures at the JM I belong to the RD Department, Carestream express request of the user. Health Rochester.

BIBLIOGRAPHY

1. Almeida MO, Phillips C, Kula K, Tulloch C. Stability of the palatal rugae as landmarks for analysis of dental casts. Angle Orthod 1995;65(1):42-48. 2. Ashmore JL, Kurland BF, King GJ, Wheeler TT, Ghafari J, Ramsay DS. A 3-dimensional analysis of molar movement during headgear treatment. Am J Orthod Dentofacial Orthop 2002;121:18-29. 3. Bailey LT, Esmailnejad A, Almeida MA. Stability of the palatal rugae as landmarks for analysis of dental casts in extraction and nonextraction cases. Angle Orthod 1996;66:73-78. 4. Bell A, Ayoub AF, Siebert P. Assessment of the accuracy of a three-dimensional imag- ing system for archiving dental study models. J Orthod 2003;30:219-223. 5. Bookstein FL. Applying landmark methods to biological outline data. In Mardia KV, Gill CA, Dryden II editors. Image fusion and shape variability. Leeds United Kingdom. University of Leeds Press. 1966:59-70. 6. Bookstein FL. Landmark methods for forms without landmarks: localizinggroup differ- ences in outline shape. Med Image Anal 1997:225-243. 7. Bookstein FL. Size and shape spaces for landmark data in two dimensions. Statistical Sciences 1986;1(2):181242. 8. Cevidanes LH, et al. Superimposition of 3D conebeam CT models of orthognathic surgery patients. Dentomaxillofac Radiol 2005;34:369-375. 9. Cevidanes LH, Styner MA, Proffit WR. Image analysis and superimposition of 3-dimensional cone-beam computed tomography models. Am J Orthod Dentofacial Orthop 2006;129:611-618. 10. Cevidanes LHC, Motta A, Proffit WR, Ackerman JL, Styynere M. Cranial base Super- imposition for 3-dimensional evaluation of soft-tissue changes. Am J Orthod Dentofa- cial Orthop 2010;137:120-129. 11. Cevidanes LHC, Figueiredo Oliveira AE, Grauer D, Styner M, Proffit WR. Clinical Ap- plication of 3D Imaging for Assessment of Treatment Outcomes. Seminars in Ortho- dontics 2011;17(1):72-80. 12. Cha BK, Lee JY, Jost-Brinkmann PG, Yoshida N. Analysis of tooth movement in ex- traction cases using three-dimensional reverse engineering technology. Eur J Orthod 2007;29:325-331. 13. Champagne M. Reliability of measurements from photocopies of study models. J Clin Orthod 1992;26:648650. 14. Choi JI, Cha BK, Jost-Brinkmann PG, Choi DS, Jang IS. Validity of palatal superimpo- sition of 3-dimensional digital models in cases treated with rapid maxillary expansion and maxillary protraction headgear. Korean J Orthod 2012;42:235-241. 15. Dryden and Mardia Statistical Shape Analysis. Chapters 1,2,3,4. 1998.

20 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence 3D cephalometry and artificial intelligence

16. Faure J, Treil J, Borianne Ph, Casteigt J, Baron P. Céphalométrie tridimensionnelle en orthopédie dento-faciale. Les possibilités actuelles du logiciel Cépha 3Dt. Orthod Fr 2002;74:19-37. 1 7. Faure J, Baron P, Treil J. Analyse céphalométrique tridimensionnelle: diagnostic des dysmorphies antéro-postérieures et verticales. Orthod Fr 2005;76(2):91110. 18. Faure J, Marshal-Sixou Ch, Dah-Jouonzo H, Oueiss A, Baron P. Analyse céphalométri- que tridimensionnelle à 14 points: compensation des dysmorphies maxillo-faciales. Orthod Fr 2008.79/251-261. 19. Faure J, Oueiss A, Marchal-Sixou Ch, Braga J, Treil J. Céphalométrie tridimension- nelle: applications en clinique et en recherche (Rapport du Congrès SFODFSIDO Venise). Orthod Fr 2008;79:13-30. 20. Faure J, Oueiss A, Treil J, Braga J. Étude en morphométrie géométrique des asymétries faciales: rapports avec les formes basicrâniennes et les pathologies maxillo-faciales ou occlusales. Rev Orthop Dento Faciale 2013;47:155-171. 21. Hoefert CS, et al. Implementing a Superimposition and Measurement Model for 3D Sagittal of Therapyinduced Changes in Facial Soft Tissue: a Pilot Study. J Orofac Or- thop 2010;3:221-234. 22. Hoggan BR, Sadowsky C. The use of palatal rugae for the assessment of anteropos- terior tooth movements. Am J Orthod Dentofacial Orthop 2001;119:482-8. 23. Jang I, et al. A novel method for the assessment of three-dimensional tooth move- ment during orthodontic treatment. Angle Orthod 2009;79:447-453. 24. Kau CH, et al. The 3-Dimensional Construction of the Average 11-Year-Old Child Face: A Clinical Evaluation and Application. J Oral Maxillofac Surg 2006; 64:10861092. 25. Kau CH, Richmond S, Incaprera AK, English J, Kia JJ. Three dimensional surface ac- quisition systems for the study of facial morphology and their application to maxillo- facial surgery. Int J Med Robotics Comput Assist Surg 2007;3:97-110. 26. Kau CH, Richmond S. Three Diiemnsional analysis of facial morphology surface chang- es in untreated children from 12 to 14 years of age. Am J Orthod Dentofacial Orthop 2008;134:751-760. 27. Kau CH, et al. Use of 3-dimensional surface acquisition to study facial morphology in 5populations. Am J Orthod Dentofaical Orthop 2010;137: S56.e1-S56.e9. 28. Kau CH. Creation of the virtual patient for the study of facial morphology. Facial Plast Surg Clin North Am 2011;19(4):615-622. 29. Keating AP, Knox J, Bibb R, Zhurov AI. A comparison of plaster, digital and recon- structed study model accuracy. J Orthod 2008;35:191-201. 30. Lagravere MO, et al. Reliability of traditional cephalometric landmarks as seen in three dimensional analysis in maxillary expansion treatments. Angle Orthod 2009;79(6):1047-1056. 31. Luebbert J, Ghoneima A, Lagravere MO. Effets squelettiques et dentaires de l’expan- sion maxillaire rapide évaluée par l’imagerie tridimensionnelle: une étude multicentri- que. International Orthodontics 2016;14:1531. 32. Maal TJ, Plooij JM, Rangel FA, Mullemans W, Schutyser FA, Berge SJ. The accuracy of matching three dimensional photographs with skin surfaces derived from cone beam computed tomography. Int J Maxillofac Surg 2008;37:641-646.

J Dentofacial Anom Orthod 2016;19:409 21 J. FAURE, A. OUEISS, J. TREIL, S. CHEN, V. WONG, J.-M. INGLESE

33. Morenhout BA, Gelaude F, Swennen GR, Casselman JW, Van der Sloten J, Mom- maerts MY. Accuracy and repeatability of cone beam computed tomography (CBCT) measurements used in the determination of facial indices in the laboratory setup. J CranioMaxillofacial Surg 2009;37:18-23. 34. Nanda V, et al. Quantitative analysis of 3-dimensional facial soft tissue photograph- ic images: technical methods and clinical application. Progress in Orthodontics 2015;16:21. 35. Olszewski R, Tanesy O, Cosnard G, Reychler H. Reproductibility of osseous land- marks used for computed tomography based three-dimensional cephalometric anal- yses. J Craniomaxillofac Surg 2010;38(3):214-221. 36. Oueiss A. Les rapports tridimensionnels de la base du crâne et du massif maxillo-­ facial: intérêt en Orthodontie et en Anthropologie. Thèse DU.8/04/2010.UPS Toulouse III. 37. Oueiss A, Pages C, Treil J, Braga J, Baron P, Faure J. Étude des asymétries dans les grandes dysmorphies antéro-postérieures. Orthotod Fr 2010;81:235-244. 38. Raghav P, Suri R, Reddy M, Phull T, Rana T, Jain S. Stability of the palatine rugae: an adjunct to forensic identification. J Dent Health Sci 2015;2(1):39-48. 39. Rossini G, Parrini S, Castroflorio T, Deregibus A, Debernardid CL. Diagnostic accuracy and measurement sensitivity of digital models for orthodontic purposes: A systemat- ic review. Am J of Orthd Dentofacial Orthop 2016;49(2):161-170. 40. Seagara DC, et al. Morphology of an adult Egyptian Population and an adult Housto- nian White population Compared Using 3D Imaging. Angle Orthod 2009;79:991-999. 41. Simmons JD, Moore RN, Erickson LC. A longitudinal study of anteroposterior growth changes in the palatine rugae. J Dent Res 1987;66:1512-1515. 42. Treil J, Casteigt J, Escude B, Roch P, Puech C. Intérêt de l’imagerie 3D dans le bilan pré-chirurgical des dysmorphoses maxillo-faciales. Radiologie J CEPUR 1992;12:20-25. 43. Treil J, Casteigt J, Jaeger M, Cavezian R. La charpente maxillo-andibulaire: nouvelle ap- proche cranio-faciométrique tridimensionnelle. Actual Odontostomatol 1993;188627-37. 44. Treil J, Casteigt J, Roch P, Puech C, Cavezian R. Imagerie 3D des dysmorphoses max- illo-faciales. Orthod Fr 1994;64:295-308. 45. Treil J, Casteigt J, Madrid C, Borianne Ph. Une nouvelle construction céphalométrique tridimensionnelle. Un nouveau paramétrage d’analyse tridimensionnelle: les axes d’inertie. Un nouveau concept de l’équilibre maxillo-facial. Orthod Fr 1997;68:171-181. 46. Treil J, Faure J, Braga J, Casteigt J, Borianne Ph. Imagerie and céphalométrie tridi- mensionnelles des asymétries cranio-faciales. Orthod Fr 2002;73:179-197. 47. Treil J, Casteigt, Faure J, Horn A, Horn A, Labadie M-P. Orthodontie et tomodensi- tométrie. Rev Othop Dento Faciale 2003;37:9-21. 48. Zamora N, Llamas JM, Cibrin R, Gandia JL, Paredes V. A study of reproducibility of cephalometric landmarks when undertaking a three dimensional (3D) . Med Oral Patol Oral Cir Bucal 2012;17(4): 678-688.

22 Faure J., Oueiss A., Treil J., Chen S., Wong V., Inglese J.-M. 3D cephalometry and artificial intelligence