"Three Dimensional Imaging Based Diagnosis for Obstructive Sleep Apnoea: A Conceptual Framework 1 1 2 Syed M. S. Islam, Mithran S. Goonewardene and Paul Sillifant 1 School of Dentistry, The University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia 2 Department of Oral and Maxillofacial Surgery, Royal Perth Hospital, 197 Wellington Street, Perth, WA 6000, Australia 1 2 fsyed.islam,[email protected], [email protected] Abstract. ObstructiveSleepApnoea(OSA)isadisorderinwhichrepet- itiveperiodiccessationofbreathingfor10secondsormoreoccursduring sleep despite increased e?ort to breathe. It leads to day-time sleepiness, poorerhealth,increasedhealthcareandhigherwork-relatedandroadac- cidentscostingthenationaleconomybillionsofdollarsperyear.Earlyin- tervention may improve health outcomes for the su?erers. In this article, a hierarchical diagnostic approach is proposed in which at ?rst a quick and safe three-dimensional (3D) surface imaging based technique is used to identify patients susceptible to OSA, thereby allowing a cost-e?ective patient screening. The susceptible patients are referred for volume imag- ing such as Cone Beam Computed Tomography (CBCT) from which the airway and other hard-tissue anatomical features can be extracted. Age and gender speci?c 3D facial norms and di?erent thresholds have been proposedtocomputeagainstwhichindividualizedfeaturescanbejudged to determine the presence of OSA. Finally, the severity of OSA is mea- sured by polysomnography sleep study only for those patients who are con?rmed for OSA by both surface and volume image-based analysis. 1 Introduction Sleep apnoea is a serious health issue with signi?cant public health implications [6, 13]. There are three types of sleep apnoea: obstructive (OSA), central (CSA) and mixed (combination of the two). In OSA (84% of cases), mechanical factors play an integral role in the reduction of air?ow despite continued respiratory e?ort [1]. In CSA (0.4% of cases) the physiological respiratory control processes fail to maintain the required respiratory function for optimal health. OSA is characterised by the presence of apnoeas (i.e. a complete cessation of breathing despite respiratory e?ort) or hypopneas, de?ned as greater than 30% reduction in chest and/or abdominal expansion during breathing or shal- low breathing lasting at least 10 seconds combined with at least a 4% reduction in oxygen desaturation. Numerous indices have been developed to express the 55severity of sleep apnoea diagnosed using polysomnography and include the ap- noea index (AI) which represents the total number of apnoeas per hour and the Apnoea-Hypopnea Index (AHI), which represents the total combined apnoeas and hypopnoeas per hour. The AHI has been divided into severity scales: mild (5< AHI< 15), moderate (15< AHI< 30) and severe (AHI> 30). Additional indices that have been utilised include sleep arousals (Respiratory Disturbance Index) and subjective patient perceptions of sleep impact on daytime activities (Epworth Sleepiness Scale). During apnoeic episodes, arterial blood oxygen saturation decreases, and sympathetic activity and blood pressure increases. Each apnoeic episode ends with an arousal from sleep, resulting in marked fragmentation of sleep in af- fected individuals. Excessive daytime sleepiness is a major consequence of OSA. OSA has also been linked to signi?cant conditions such as hypertension [16, 9], ischaemicheartdiseaseandstroke[18],prematuredeath[17],andimpairmentof cognitive functions [8] which may contribute to motor vehicle and workplace re- lated accidents (comparable to functioning while intoxicated) [7]. A study from The University of British Columbia demonstrated that a person with OSA is twice as likely to be involved in a motor vehicle accident [19]. For untreated individuals, it has been established that there is a 37% higher 5-year morbidity and mortality rate [14]. It is estimated that 775,000 Australians (4.7% of the adult population) su?er from OSA [15]. The Busselton (Australia) Health Survey [2] of 294 men aged 40 to 65 years revealed that about 26% of individuals have mild and 10% have severe levels of sleep apnoea. The total ?nancial and non-?nancial burden of OSA in Australia was estimated as 21.2 billion dollars in 2010 including direct health care cost of $575.42 million and indirect health care cost (due to lost productivity,deadweight loss, workplace/motorvehicleaccidents,social security payments etc.) of $2.6 billion [18]. In U.S. it was estimated in 2008 that the averageadditionalannualhealthcarecostofanuntreatedsleepapnoeapatientis US$1,336contributinganestimatedtotalof$3.4billion/yearadditionalmedical costs [1]. In this article, we introduce a novel quantitative diagnostic method for OSA based on the combination of two approaches related to two di?erent imaging modalities (surface and volume). The ?rst approach is based on the analysis of a three-dimensional surface scan of a subject (using e.g. a 3dMD face scanner). We propose to extract quantitative facial features from the scan to di?erentiate between facial morphologies of OSA patients and normal non-apnoeic individ- uals. The relative position of the upper and lower jaws to the skull base and in turn to each other can be assessed as represented by the external facial ap- pearance. These facial features can be evaluated to determine the relationship between facial morphology and the severity of OSA. 3D surface facial scanning has the advantage of being a non-invasive imaging tool which does not require exposure to ionizing radiation. The second approach relates to the application of state-of-the-art dental imaging in the form of a Cone Beam CT to obtain a 3D (volumetric) representation of the hard and soft tissues. The determination 56of the morphology (shape and structure) of the airway of OSA patients should help in revealing any signi?cant deviations from the airway of normal individ- uals. As Cone Beam CT is a readily available imaging tool in most clinics, the proposed diagnostic method is easily accessible with many control non-OSA pa- tients imaged for unrelated dental anomalies. The overall outcome of the article is the development of improved conservative diagnostic methods which will be accessible to wider patient groups and will contribute in early intervention. The rest of the article is organized as follows. Various approaches currently used for the diagnosis of OSA is described in Section 2. The conceptual frame- work for our proposed approach is elaborated in Section 3. Proposal for the evaluation of the new diagnostic method is discussed in Section 4 followed by the conclusions in Section 5. 2 Existing Diagnostic Approaches for OSA OSA is seen more frequently in older males and is related to many predisposing factors such as increased Body Mass Index (BMI), increased neckcircumference, smoking, alcohol consumption and enlarged tonsils and adenoids. Clinicians also recognise speci?c dentofacial deformities which predispose individuals to the development of OSA. The obvious retrusion or underdevelopment of the lower jaw and and/or the upper jaw alerts the clinician to the possibility of a patient susceptible to OSA. Today, overnight polysomnography remains the ‘gold standard’ diagnostic methodforOSA.Itisamonitoredsleepstudytorecordbiophysiologicalchanges thatoccurduringsleep.Measurementsincludeelectroencephalogram,electroocu- lograms, submental electromyogram, oronasal air?ow, chest wall motion, and arterial oxygen saturation. In addition to the signi?cant inconvenience to the patient, polysomnography requires sophisticated specialist facilities, technical and scienti?c sta? and sleep clinicians, which are commonly not available in all regions. Imaging techniques have been considered as useful adjunctive tools to diag- nose and plan the treatment of OSA, with the radiographic head ?lm (cephalo- metric) analysis being the most convenient and widely used [3]. However, the cephalometricanalysisisinherentlylimitedbecauseofitstwodimensionalimag- ing and the lack of information about the airway volume and dimensions [5]. In addition, measurements are obtained with the patient in the upright position which may not accurately re?ect the distortion of the airway in the supine sleep- ing position. This may create an underestimation of the degree and pattern of airway narrowing and/or collapse. Lee et al. [10, 11, 12] analysed facial char- acteristics to predict OSA with an accuracy of 76.1% using 2D photographic and cephalometric images. These have limitations compared to 3D surface and volume data. For example, while they demonstrated a relationship between fa- cial structural measurements such as alar width and intercanthal distance, they did not assess 3D positional relationships of the relevant structural components representing the underlying jaw base, which is the focus of this article. 57During the last few years, there has been signi?cant interest in developing conservative,cost-e?ective,patient-convenientandwidelyapplicablemethodsto diagnose and treat OSA. Although the morphology of patients diagnosed with OSA has been well documented using two dimensional (2D) imaging techniques, and to a much lesser degree using 3D imaging techniques, no speci?c strati- ?ed evaluation has demonstrated the impact of progressive distortions of the maxillomandibular structures on air?ow and sleep performance. 3 Proposed Methods and Techniques Considering the cost e?ectiveness and the simplicity, we propose a hierarchical framework for diagnosing OSA. We would like to keep the cheaper and widely accessiblemeasuresatthebeginningandthusscreeningoutanumberofpatients before suggesting for more expensive and exhaustive approaches. The detailed framework is described in this section. 3.1 Statistical Design A null hypothesis for developing the new diagnostic approach can be de?ned as follows: there will be a statistically signi?cant di?erence in the proportion of patients who are correctly diagnosed with OSA using the new method as compared to the gold standard. The sample size for the above hypothesis can conservatively be estimated using an expected sensitivity (probability of correctly identifying a patient as positive by the proposed approach given they have OSA) of 0.85 and speci?city (probability of correctly identifying a patient as negative by the new approach given they do not have OSA) of 0.95, and a 95% con?dence level. A sample size of 100 OSA patients and 100 non-OSA participants would provide a 0.07 precision for sensitivity and 0.04 precision for speci?city. 3.2 Determination of Norms and Thresholds This approach requires a prior set up of age and gender speci?c facial norms (nn) used as references. For that purpose, we propose to compute the age and gender speci?c average faces from a large sample of non-OSA subjects. In ad- dition to these average-face norms, we also propose to determine some other thresholds associated with other discriminating features as illustrated in Fig. 1 and explained below. Threshold t can be established as follows from 3D ear to ear facial surface 1 images(e.g.Fig.2)ofthe100patientsdiagnosedwithOSAbypolysomnography. The face area can be detected and cropped and various surface features (e.g. length of the maxilla, mandible and chin and the circumference of the neck) can be extracted. The relative shape ratios (RSRs) of these di?erent features (e.g. length of maxilla with respect to the mandible and that of maxilla and mandible compared to the forehead and neck) can be computed. These features then can 58Age and genderspecific normsFind the3D surface (nn) Determinemostimage age andcommonDetect1 gender t 1 Measuredeviationsandspecificdifferentin OSAextractdeviationssurfacen patientsear to earFacialfeaturesface data Compute relativeimages ofshape ratiospersons(RSR) diagnosedwith OSA Find RSRs3Dthat arevolumetric Measure Determine mostlyimage1 Segmentvolumetric correlation related toairway,parameters between airwayn mandible t 2 morpholo of the airwayandanatomic morphology gy ofmaxillacomponents and RSRs most ofthe OSApatients Find theComputeCompute consistentmaxilla- Soft-tissue factor ofmandiblet 3 compen- soft-tissuerelativesation compen- shape ratiossation Compute age and gender specifict 4 average morphology of the airway Fig.1.Blockdiagramofthecomputationofdi?erentthresholds(t ; t ; t ,and t )used 1 2 3 4 in the proposed diagnostic algorithm. Fig.2. 3D textured image of a person’s frontal (left) and right pro?le. 59be compared with the age and gender speci?c norms to outline any deviations from the norms. The threshold t can then be derived from these deviations. 1 Three more thresholds can be determined from 3D volumetric images which can be acquired using a Cone Beam CT scanner from the same patients above. The volumetric data of the airway (Fig. 3) and other anatomical features can be segmented from these data using commercial software such as Dolphin, 3dMD- vultus and 3D Slicer. Di?erent volumetric parameters can be measured and statistically correlated with the facial RSRs computed from the facial surface images. The RSR (of each age and gender group) with the highest correlation factor can be used as a threshold (t ). The relative shape ratio of maxilla and 2 mandible computed from volumetric data can be compared with those obtained from surface data (3dMD) to evaluate the most common soft-tissue compensa- tion factor (t ). The average morphology of the airway (threshold, t ) of the 3 4 di?erent age and gender subgroups can be computed using the above software or computer programming using MATLAB. Hard-tissue Airway Soft-tissue Fig.3. 3D volumetric image of an OSA patient and his digitally segmented airway represented in wireframe model. 3.3 Diagnosis Using the New Approach AsillustratedinFig.4,intheproposeddiagnosticframework,asubjectpresent- ingforanOSAtestwill?rstlybediagnosedusingasurfaceimage.A3dMDscan (e.g. Fig. 2) will be taken using the 3dMD Facial Scan System. The captured image data will be represented as a 3D surface mesh. Then quantitative facial shape features and ratios will be extracted or derived from the surface data. An individualized norm will be determined based on the age and gender spe- ci?c norms (nn) to localize and quantify any shape deviations (d ) of the facial 1 60Age and genderspecific average Determine Computeface (nn) individualized relative shapenorm ratio (RSR) Localize3D surfaceDetect and Measureandimage extract 2D different(d +t -t )+(RSR- 1 3 1 quantifyand 3D ear to shape t )>=t 2 5 deviationsear face data features A patient(d ) 1 Noapproachi Yes ng forOSA test MeasureLocalize andSegment airwayquantifyairway volumetricVolume deviations (d ) 2 parameters image Average airway norm of theOSA patients (t ) 4 NoYes Nod >=t 2 6 OSA Perform polysomnographyand other clinical observations NoAHI>=5 Yes OSA Fig.4. Block diagram of the proposed diagnostic methods. 61"