www.nature.com/scientificreports scientific reports OPEN coe for pten Automated caries detection in vivo using a 3D intraoral scanner Stavroula Michou'?, Mathias S. Lambach**, Panagiotis Ntovas*$, Ana R. Benetti!, ‘Azam Bakhshandeh*, Christos Rahiotis®, Kim R. Ekstrand? & Christoph Vannahme? ‘The use of 3D intraoral scanners (10S) and software that can support automated detection and tive monitoring of oral diseases such as caries, tooth wear or periodontal diseases, is increasingly 1g attention from researchers and industry. This study clinically validates an automated caries scoring system for occlusal caries detection and classification, previously defined for an IOS system featuring fluorescence (TRIOS 4, 3Shape TRIOS A/S, Denmark). Four algorithms (ALG4, ALG2, ALG3, ALG4) are assessed for the 10S; the first three are based only on fluorescence infermation, while ALG4 10 account the tooth color information, The diagnostic performance of these automated algorithms is compared with the diagnostic performance of the clinical visual examination, while | assessment is used as reference. Additionally, possible differences between in vitro and nostic performance of the IOS system are investigated. The algorithms show comparable nostic performance to the visual examination with no significant difference in the area under the ROC curves (p > 0.05). Only minor differences between theirin vitro and in vivo diagnostic performance are noted but no significant differences in the area under the ROC curves, (p > 0.05). novel IOS system exhibits encouraging performance for clinical application on occlusal caries detection and classification. Different approaches can be investigated for possible optimization of the system. ‘he use of 3D intraoral scanners (10S) and corresponding software for oral disease detection and monitoring has proven potential". There is increasing development inthis area, bth by companies that produce medical devices, and by researchers secking improved devices and software that can support automated detection and objective monitoring of oral diseases such a caries, tooth wear and periodontal diseases either in a clinical setup or remotely" The implementation and diagnostic performance ofan autonsated caries scoring system based on the fluo rescence method using blue-violet light (415 nm wavelength) in « SD IOS system (TRIOS 3, 3Shape TRIOS A/S, Denmark) has previously been investigated Different caries classification algorithms were investigated for this luorescence-hased 10S system, showing good in vitro diagnostic performance when assessing occlusal caties lesions. More specifically, at the caries stages where the comparison between the conventional methods tnd the [OS caries classification system was possible, the best-performing IOS algorithms employing optimal ‘cut-offs showed slightly higher sum of Sensitivity (SE) and Specificity (SP) (SE*SP ~ 1.58-1.84) than the visual: tactile examination (SE*SP = 1,73-1.81) and significantly higher than the radiographic examination (SE+SP = 137-178), The only exception was observed for the caries lesions located inthe outer tied of dentin, where the radiographic assessment shoved a higher value (SE+SP = 1.78) compared tothe best-performing 10S algorithms (SE+SP = 1.67-1.69). In that study, both visual-tactle and radiographic assessments employed the International Caries Detection and Classification System criteria (ICDAS)". The idea behind the development of the auto- sated caries detection and classification system to accompany the 3D IOS is that by combining a method similar tothe well-documented Quantitative Light-Induced Fluorescence (QLE)"'*"" withthe 3D information provided by the 10S, detection and monitoring of the caries lesions can potentially he improved. ‘The study mentioned above! and later investigations led to the definition of different algorithms for an automated caries scoring system, and its implementation in a prototype software accompanying the 3D IOS (TRIOS 4, 38hape TRIOS A/S, Denmark). These algorithms employ red and green fluorescence (Ray Go) signal! resulting from scanning with light at 415 nm, However, tooth color (Red, Green and Blue) signal (RG, ‘simultancnusly obtained from scanning th teeth with wit light. Its speculated that by combining all the "Department of Odontology, University of Copenhagen, 2200 Copenhagen, Denmark *3Shape TRIOS A/S, 11060 Copenhagen, Denmark. "School of Dentistry National nd Kapodistian University of Athens, 13527 Athens, Grnece, "These authors contibuted equally: Mathias §, Lambach and Panagiotis Ntovas. email: stmi@ sund ku dk Scientific Reports| (2022) [reps dei ergo 1038)s42598.021-00259.»/ natureportfoliowww.nature.com/scientificreports! valle color information on a 3D model, and by analyzing any difference of color signal intensity on the tooth surface together with the fluorescence changes corresponding to sound and demineralized dental tissue, the accuracy in detecting occlusal cares lesions could be increased!'=". This hypothesis is supported by previous research, in which similar approaches combining the fuorescence method with reflectance enhancement shoseed relatively accurate detection and monitoring of caries lesions in vitro (SESP = 1.55)'*”. Thus, an algorithm combining all the color information on the 3D model (Ryo, Guy, G, B) was defined and tested on existing Sample. This specific algorithm showed the best in vitro diagnostic performance for oclusal caries detection and classification at one optimal cut-off in enamel and two in dentin (area under the ROC curve, Az = 08, SE ~ O.83and SP» 0187), which motivated us to include it in this validation stad: Based on the diagnostic performance of the 3D TOS for in vitro occlusal caries detection! and considering the unique advantage of 3D models, which combine geometry, color signal from the tissues, and inthis case, fluorescence signal we assume tha this device can help to overcome some limations observed fr the existing 2D intraoral cameras featuring fluorescence for caries detection. For example, difficulties in obtaining repro: ucible 2D intraoral images for monitoring cares lesions overtime is a common issue, limited largely bythe {mage acquisition angle. The later can significantly affect the siz ofthe lesion depicted on the 2D images" but is expected to have less influence on the assessment using 3D models where the averaging of image data gives less noise and eliminates images obtained feom steep angles Despite the good results obtained forthe TOS system in vitr!, i was essential to validate the defined algo- rithms and corresponding cut-off on « new blind sample in vivo”, Previous studies assessing other devices featuring fluorescence for caties detection have observed significant diferences among the devices in viteo ding- nostic performance at optimal cutoffs and ther subsequent performance achieved in in vivo validation studies, where pre-defined cutoffs wee assessed on independent samples", The latter has led previous researchers 10 the conclusion thatthe in vitro defined cut-offs noed modifiation for in vivo application. Aim. The purpose of this study was to clinically validate four automated caries scoring system algorithms previously defined for the IOS system, using histological assessment as reference method. Further aims were: {G)to compare the performance ofthe automated scoring system with the clinical examination employing the ICDAS criteria; and (i) to assess posible differences in the performance of the automated system under in Vitro and in vivo conditions, Materials and methods Study sample. Sample size calculation was done using the formal described by Burderer, fora confidence interval at 95% absolute eror a 01, and based on the expected dagnostc performance forthe 1OS system (SE 2 084, SP20.76)- These values were based onthe performance ofthe investigated device the iterate This Calculation sulted ina minim af 100 examination sites tat shoul be icluded i the current study Permanent molars and premolars scheduled for extraction athe surgery depariment of the School of Den tisryof the Univesity of Copenhagen were considered for inclusion in the study. The age range of patents ‘eas fom 18 to 60 years ol. Teeth with severe developmental defects, calculus onthe occlusal surface, vine tensive cares lesions on other surfaces than the oclua and restored teeth were not include in the sample ‘According to these criteria, 58 teeth scheduled fr extraction were selected for examination. Ethics. This clinical study received ethical approval from the Research thies Commitee of the Schoo! of Dentistry ofthe National and Kapodistrian University of Athens, Greece (prot. nt 423/08.07.2019). The study was conducted in accordance with the declaration of Helsinki and the General Data Protection Regulation (GDPR). All clinical steps and scanning of the extracted teth were conducted atthe aforementioned Univer sity. Thereafter, all extracted teth were fully anonymized and sent for histological analysis to the University of Copenhagen, Denmark, According to the rues in Denmark, research projects involving completely anonymous cr anonymized human biological material, whichis collected in accordance withthe legislation tthe collection Sit. are exempted from notification to the Danish Committee system (ef, Article 14(3) ofthe Committes ACD. All study participants gave informed consent and agreed to publish anonymized information or images in an online publication, Study design. The overall study workllow is presented in Fig. This in vivo study with ia vitro validation assessed four diferent algorithms (ALGI-ALGA) implemented in the IOS system for automated caries detection and classification. 3D models of the examined teth were obtained both in vivo and in ito, ie before and after tooth extraction, in onder to assess any possible ferences in the algorithms performance in different conditions. The ater could potentially help dra some conclusions regard ing the validity ofthe in vitro cares detection results obtained for this 3D 10S system, and the in vive applicability of in vitro results. Additionally, a visual-tactle examination using the ICDAS criteria was conducted ia vivo and histological assessment was used as reference test in viteo (Table 1. information regarding the examiners) calibration and blinding are provided in the supplementary material Visual examination (ICDAS). the clinical examiner (PN.) defined one to three examination sites in the ‘occlusal pits and fissures ofeach selected tooth and examined al teeth in vivo employing the visual ICDAS eri teria for caries classification", Examination was performed on dry surface, under proper illumination and alier polishing ofthe occlusal surfaces with prophylactic brushes and a low-speed handpiece (Kavo Intra 20k) ‘One score (ICDASO-ICDAS6) was assigned to each examination site, and affer the 3D model aquisition, the exact position ofthe examination ste was annotated on the 3D model Fig. Iv Scientific Reports) (onsyansn76 | beeps idl org/L01036)541598-021.00259.». nature portfoliowww.nature.com/scientificreports! Methods overview 1) visual examination (ICDAS) ii) Intraoral 3D scanning Iv) Automated caries scoring system (ALG4) Tncicaton of caries stage i Moderate extensive (02-05) i) Extreorl 30 scanning rvai e105 Cinaucien scan ¥) Examination annotation vi) Histology Figure 1. Study methods overview in vivo and in vito. 3D models ofthe same tooth seanned (ja) in vivo and (iia) in vitro using white light; the tooth color signal was mapped onto the models. The same tooth was seanned (Gib) in vivo and (ib) in vitro with the 415 nm wavelength ight, which excites fluorescence from the dental tissues. (iv) caries score indication based on TRIOS patient monitoring software (3Shape TRIOS A/S, Denmask) according to ALGA. Indication of caries stages Initial caries lesions in enamel and outer thied of dentin (Histology E1-D1), Moderate-extensve; caries lesions in middle and inner thirds of dentin (Histology D2~ 3) Insuificient sean: insufficient data on tooth olor and/or florescence that dacs not allow the automated. caries sore calculation. (x) selected examination sits (ab) annotated on the 3D model. (vi tooth sectioning lines corresponding to the examination sites (ab) forthe histological assessment, On the tooth section, the red measurement line corresponds tothe demineralization depth and the blue measurement line corresponds (othe ‘enamel thickness 3D scanning, Subsequent to visual examination, all teeth were scanned in vivo using the 3D 10S TRIOS 4 (Shape TRIOS A/S, Denmark) aided by commercial software (TRIOS vers 1.182.11 and Dental Desktop vers. 16.8.1, 3Shape TRIOS A/S, Denmark) and aecording to the manufacturers recommendations: the dental lamp \was switched off other external ight was imited as much as possible, teth sacs were clean and dry and the recommended scanning strategy was followed. First, by scanning with white light, a digital 3D model ofthe teeth with tooth color texture was created (Fig 1). Then, by scanning a second time using light at 415 nm, a texture representing the fluorescence signal received from the issues was mapped onto the 3D model (Fig ib), The intraoral scanning procedure was final: ized when sufficient tooth color and fluorescence information was abtained according to the software indication, Following in vivo intraoral scanning, the teth were extracted and transferred shorty thereafter to the labora tory for in vitro scanning, There the teeth were mounted on individual bases made of putty impression mate rial (Zetalabor, Zhermack, Italy) and scanned again with the same [OS system, following the same procedures described forthe intraoral scanning in vive. The in vitro models were obtained in a dark room (Fig, lia), ‘within 48 hours from tooth extraction. Intraoral scanner’s algorithms. _Fourdilferentalgorthms (ALGI-ALGA) defined for cares detection and classification on the 3D models were assessed, An article describing the definition ofthe frst three algorithms Scientific Reports (2023)128276 | beeps idl org/L01036)541598-021.00259.». natureportfoliowww.nature.com/scientificreports! Sound ooh sie show 0 We SOUND | Ps Sound Sound tesound ‘Enercon sewed afer ming snd ster Sconds of ang 1: Fist vial Gunn ena opal f te Cars in th ater aff ename rors) ries theemrantsof por Fore fers iene {Seas oer aber ccs tidying Car ati ier al of rane —tall 2: Dati visa change in ml (opcy ng deta jaetion DED Se dsclrtion wah wn tht sad ‘Cari inthe outer thi of dentin 2e Cais inthe er hid of Senin ig diag A white o bow ptlsion wih oad ‘ose Non cited sure wih an undying DESTIN 2 Casi ein ‘Satin shad: wich obvious origned on {Cakes in th loans thio di |e se Bing erate 5: Visual dnt avi page ode ‘Sure enamel and exon dei. Fons moet aloft res) {hu ual dint aay wahesposed sea DbeCais inthe mie hin of ein Cai inthe ane tid data ‘Table 1. Scoring systems employed by the different methods according to histology. (ALGI-ALG3) was published previously by Michou et al!, Mathematical functions fe fy the mentioned study correspond to ALGI-ALG3 in the current study. The last algorithm, ALG, was defined at a later stage using the same sample and methods as deseribed in the above mentioned study". Histology was used asthe reference ‘method for the definition ofall algorithms. Receiver Opersting Characteristic (ROC) analyses were conducted ‘on the raw data from each algorithm. Optimal cut-ofs fr different cares severity levels according to histology ‘were defined by the sum of SE and SP at each histological level Table |), For ALGI and ALG2, reliable independent cut-offs (SE+SP > 1.7) could only be defined for two caries sever ity level: (i) caries lesions in enamel (1) and (ji) caries lesions in dentin (= D1). For ALG3 and ALGA, an additional cut-off corresponding to (i) caries lesions in the middle-inner third of dentin (2 D2) was also defined, Thus, using ALGS and ALGA the lesions in the outer third of dentin received a different score than the lesions inthe middie-inner third (Table 1). Thefts three algorithms (ALGI-ALG3) were based exclusively on the fluorescence signal received by the dental tissues. More specifically: ALGI represents the absolute green ffuorescence signal (Giys) on each examination site; ALG2 represents again the Co but taking as reference the average Gia from the sound surfaces on the same tooth; and ALGS represents both red (Ra) and green fluorescence signal (Gps) on the examination sites and uses as reference the average Ryo and Gg from sound surfaces located on the same tooth. The last algorithm, ALGA, was found by logistic regression, and takes ito account both fluorescence (Ray, Cu) and tooth color signal (R, G, B) from the examination sites using the ‘sound tooth surfaces a reference, ‘Rather than selecting the areas of interest manually in order to calculate the caries scores, inthe current study the prototype software already integrated the algorithms ALGI-ALGH. This software was based on the commercially-available TRIOS Patient Monitoring software (3Shape TRIOS A/S, Denmark) and enabled the automated display ofa color overlay on the 3D models of the teeth, which represented the caries severity indica: tion on the model according to each algorithm (Fig. liv) Using this custom-made software, an independent examiner not involved in the clinical examination (S.M.) assessed the 3D models acquired both in vivo and in vitro. The automated seores given from each algorithm on the 3D models were registered on the same examination sites initially selected by the clinical examiner (PN). “The scoring system corresponding to each algorithm is shown in Table Reference test—Histology. Histological asessment was used as the reference standard such as that described in the literature’. The asim caries lesion depth s well the enamel or dentin thickness (at the same position), were registered for each examination ste (Fig. Ii). Based en the outcome resulting from the faction caries lesion depthienamel thickness or caries lesion depthdentin thickness fr lesions located in enamel and dentin, respectively, the ellowing histological scores were given to each examination st: BO sounds = El lesions in the outer half of enamel (fractions = 0.5) = EB lesions in the inner haf of enamel including the dentin-enamel junction (DE)) (fractions > 0.5); i lesions inthe outer third of dentin (fractions < 0.33), n the middle third of dentin (fractions > 0.33 and < 0,66); and n the inner third of dentin, with or without pulp involvement (fractions > 0.66) Scientific Reports|_ (202:)1.28276 | beeps idl org/L01036)541598-021.00259.». nature portfoliowww.nature.com/scientificreports! Data analysis. ll examination sites were assigned an independent score using the different software algo rithms, visual assessment (ICDAS), and histology. ‘Spearman’ rank correlation coefficient (,) was used to assess posible cortelation between the histology and the scores originated from algorithms or visual assessment. The diagnostic performance fo all methods was expressed by ROC analyses and contingency tables using histology a reference (see Supplementary table SI). ‘Area under the ROC curve (Az), Sensitivity (SE), Specificity (SP) and accuracy (ACC) were then calculated aftr dlichotomnizing the data atthe El, D1 and D2 histological levels, which correspond to the three cut-offs defined forthe algorithms. Areas under the ROC curves for the investigated methods atthe El, D1 and D2 levels were compared pairwise using DeLong’salgorithm®, while SEand SP values were compared using McNemar’ test The standard error (Std. Eer.) for SE and SP was adjusted for possible clustering eect as muiple examination sites were selected on the same tot". The MeNemar-Bowker test was employed to assess possible differences between the in vivo and in vitro results forthe different algorithms, ‘Spearmanis rank correlation coefficient, contingency tables ROC analyses, andl McNemar’s test were per formed using IBM SPSS Statistics (Version 26, IBM Corporation). Other calculations weve performed in Excel (Microsoft Ofice 2016) and comparison of areas under ROC curves was made using MedCale statistical software (Version 19.6.4, MedCale Soltware Lid, Belgium). Confidence level was defined at 95% forall statistical tests Results (Out ofthe 58 teeth initially included for examination, 5 ether did not full the study's inclusion criteria after tooth extraction and second inspection in vitro, or were destroyed while sectioning for histological analysis. Finally, a total number of 53 teeth with L18 examination sites were included for statistical analysis. Out of those, some examination sites could not be assessed using the algorithms, ether due to insufficient sean data or algo rithm failure; the number of missing examination sits for each algorithm can be seen in the contingency tables (Supplementary table $1). According to histology. out of the total number of examination sites (n = 118), 17 ‘were sound (£0), 79 were initial cares lesions in enanvel (EL, E2),8 were lesions in the outer third of dentin (D1) and 14 were lesions located in the middle-inner third of dentin (= D2). Diagnostic performance. ‘Table 2 shows descriptive results including correlation tothe histological scones (13), Az, SE, SP and ACC forall algorithms, both in vivo and in vitro, and for visual examination in vivo. Figure 2 presents the ROC curves corresponding to algorithms and visual examination in vivo, All methods resulted in Significant correlation (r,) with histology (p = 0.001): ALG3, ALG4 and visual assessment showed moderate cor relation (01 <15=0.54) and ALGI and ALG? showed fai or weak correlation (rs = 040) Table 2. Caries detection level (Histology > EL). When assessing the ability of the diferent investigated methods to detect caries lesions in vivo an! in vitro (Histology > El), all methods resulted in similar area under the ROC ‘curves (Az) no significant diferences among the Az values of different methods were observed (p = 0.05). The highest SE and ACC were exhibited by ALGI, ALG4, and visual assessment, chile significantly lower SE was found for ALG3 (p = 0.001). However, ALG3 presented the highest SP (p = 0.05) in vivo. No significant difer ences among SP Values were observed in vitro, Caries in dentin (Histology > DY). As regards the detection and classification of caries lesions in the outer third of dentin (Histology > DI) both in vitro and in vivo, ony the 10S algorithms were assessed as there is no ICDAS score for visual examination that can seliably distinguish between lesions in enamel and intial lesions in the outer third of dentin. When assessing the 3D models acquired in vivo, no significant diference among the Az for all algorithms was detected (p > 0:08). However, regarding measurements on models obtained in vitro, ‘ALG2 resulted in significantly ower Az values than in vivo (p = 0101), ALG3 and ALG4 showed the highest Az, SP, and ACC bothi vitroand in vivo. Oa the other hand, AEGI exhibited significantly lower SP (p ~ 0.05) than allthe other algorithms, but high SE Caries in the miilesiomer third of dentin (Histology > D2). 1a the midde-inner thied of dentin (Histology > D2), only the in vivo visual scores, and those from ALGS and ALG4 wore assessed. Regarding the Ar and SE values, no significant diferences among the diferent methods were observed. Visual assesment showed the lowest ACC and SP in iv with the latter being significantly interior to the SP of ALGS and ALG. [Almost identical n viteo diagnostic performance was observed for ALGS and ALG Algorithm reproducibility in vivo vs. in vitro. No significant difference was found between in vivo and {nitro ordinal scores resulting from the 1OS algorithms (MeNemas Bowker test, p > 0.03) In alton forall algorithms and aallasessed histological levels (E1, D1, D2) no significant ciference was detected between the “Ar values obtained from in vivo oi vitro assessments (p > 0.03). Homsever as regards the caries detection evel (Histology = E1), ALG1, ALG?2, and ALGA showed higher SE {nitro than in vivo (MeNemar’s test on binary data: ALGI.2 p < 0001, ALG# p = 0.08), Discussion ‘The algorithms for automated caries detection and classification defined for the IOS system (TRIOS 4, 3Shape TRIOS AS, Denmatk) were validated against histology. This study isa significant step towards implementing an automated caries scoring system in a commercial 3D [OS system, which can ad caries detection and potentially Support caries monitoring in everyday’ clinical practice, When considering the detection and classification af Scientific Reports) (onsyansn76 | beeps idl org/L01036)541598-021.00259.». natureportfoliowww.nature.com/scientificreports! ‘az Jazuoon)* (07s005)" Jaros)" |orKaa5y* |arHon0)* aay S_[e7w0asy4# Jorwansy® Jaseo05° [arnaasy'* |oania0s)* ON Fe Jassans [oss [ovo | asH a0 [osmann® e_[arwone® [aroane? [oanwaer [oaxoaa” [NA i fosnnsoy? [07116 [asroa2y* | a76033/ SP _Jossiousie joan00? Josnomy’ |oswomy [NA ace [ot oa [087 a NA a [Na wa [eannsy® [osaana* [asoaon* = [NA a [e7anis® [oss [asxa0n* i FNS [oon [O98 [Na wa [easumy’ [oswomy* [aru ‘Az Jarmonn)* jo7¥08)* |a7HH05)" | a76N5 aay [S_[ossionsy* fosicansy* [os 005" |astaoe/ ON Fee Tossensy® [ossous)* [aria2/ | assay ae _[arwons® [aswoun® [anno [axxnanr 21 you? joara6 a7a(015/% oro) SP [oaeiouni [oso |oxaaony | osiamy ce [05s 056 [ee a a [Na Wa [oasionsy® [ossa0s* son [NA wa [oxy [oascny [Na wa [exsamy? [ossany* ‘Table 2. Descriptive results for all methods assessed in vivo (a) and in vitro (br, Spearmans rank correlation coefficient; Az, area under the ROC curve; SE, sensitvity—true positive rate SP, specificity —true negative rate; ACC, diagnostic accuracy; N/A, not available, Standard ersor is provided in parenthesis, SE and SP standaed error i adjusted for elustered data. The significant diferences within the samte row are marked ‘with capital letters following the sequence A > 8 > C, Confidence level was defined at 95% fr all statistical tests, initial (Histology > Et) and moderate-cxtensive caries lesions (Histology > D2), the 10S algorithms showed diagnostic performance comparable to visual examination using ICDAS criteria, these results are in agreement with the Iiterature assessing the QLF method. The overall diagnostic performance of the different algorithms, as indicated by the area under the ROC curve, was similar forall the ALG in vivo and the visual assessment (p = 0003). However, some statstialdiferences were observed among the SE and SP values at the different dig nostic levels (p< 0.05). ALG1 and ALGA showed similar SE and SP with the visual assessment, while ALC2 and ‘ALGS deviated, showing inferior SE at initial enamel caries esions. Additionally, the ALG3 and ALGA showed Significantly higher SP regarding the more extensive dentin caries lesions (D2), where the visual assessment resulted in an increase number of false positive indications and interior SP. However, as expected in the current study and as also seen inthe literature during validation of cut-off defined for other devices"™*, Le applying cut-offs defined ina previous study on a new sample, the diagnostic performance of the investigated algarithms was considerably inferioe to the one observed at optimal cut-offs assessed in a previous in vito study . This agrees with ther studies supporting that no absolute cut-offs can be defined for the devices featuring optical caries detection with fluorescence. The defined cutoffs ean only be used fan indication forthe relative caries lesion depth" "No significant overall difference was detected regarding the performance ofthe algorithms on the 3D models ‘obsained in vivo or in vite. This finding confirms that future cares validation studies assessing this IOS system can be conducted in vitro and provide a good indication ofthe in vivo diagnostic performance. Subsequent, caries classification cut-offs defined in vitro can potentially be applied in vivo. However, a prerequisite is that appropriate methodological procedures are followed in vitro afte the tooth extraction, , short storage period Scientific Reports) (onsyansn76 | beeps idl org/L01036)541598-021.00259.». nature portfoliowww.nature.com/scientificreports! ROC Curves ~ In vivo assesements Sensitivity ROC Curves - E1 ROC curves - D1 a ROC Curves - D2 Bod od aust & ALG2 —~ALG3 of 02 —ALGa Visual examination Reference line 3s aa ae a ia ea as a a io ‘of —a5 5 —t0 1 - Specificity 1 - Specificity 1 - Specificity Figure 2._ ROC curves corresponding tothe four algorithms (ALG1-ALG#) and visual assessment investigated in vivo at the histologial levels E1, D1, D2. On the ROC curves, the Sensitivity values are plotted against the [Specificity values. The curves closer to the top-left commer indicate a etter diagnostic performance, while ccurves loser tothe reference diagonal line indicate inferior diagnostic performance, in liquid or freezing of teth to avoid the diffusion of porphyrins in the storage solution" uorescence image acquisition in a dark room to avoid the effect of external light ‘Considering the level of subjectivity involved inthe visual tamination, and the documented influence ofthe individual examiners professional experience on ts outcome”, the inclusion of only one examiner inthis study ‘might have introduced a level of bias. Significant discrepancies in this study's results Were likely to have been ‘observed if more examiners, ether with a different experience in Carilogy research or general practitioners had also conducted the visual examination Since the diagnostic performance ofthe visual asessment employing the ICDAS criterias widely investigated inthe literature, and the examiner variability is well known, thisaspect was snot addressed inthe present stud. In fact, substantial level of reproducibility and accuracy s reported when calibrated, well-trained examiners employ the ICDAS criteria such asin the current study", and therefore, based on consensus, one trained and calibrated examiner can reliably conduct the visual examination alone". ‘Some limitations are identified regarding the sample inthis study. The fact that the investigated teeth were scheduled for extraction means thatthe majority were third molars, in some cases semi-erupted, or with large ities, In addition, a few teeth were extracted for orthodontic reasons, or due to periodontal problems, The sutomated caries detection system is mainly intended to be used on permanent posterior teeth with initial to moderate caries esions for which monitoring can evaluate the progression or regression ofthe lesions as well as. the effectiveness of preventive measures. Thus, the constitution ofthe sample inthe current study was not fully representative ofthe tecth ina clinical scenario, Nevertheless, tis isan inherent limitation that, due to ethical considerations, could not be avoided ina validation study like this, where extraction and in vitro bistological assessment were caried out Furthermore, the inclusion of third molars and semi-impacted teeth of limited clinical acces created add ‘ional limitations, for example insufficient cleaning ofthe occlusal surface in some cases. Considering thatthe dental biofilm can emit strong red-orange fluorescence signal”, good cleaning of tecth is essential to assess caries lesions with the luorescence method, Otherwise, the fuorescence signal from bacteria might lead to false indications by devices assessing fluorescence. This pheniomenon became apparent in the current study as differ ences were observed in red fluorescence signal when assessing the scans ofthe same teeth obtained in vitroand in vivo (Fig. Lib versus ib. This variation in fluorescence signal resulted in higher SE values in vitr for the El histological level (ALGI, ALG2, ALG). Additionally in some cases, de to limited access ofthe scanner to third molars in vivo, aceas of insufficient 3D scanning data (Fig. 19) on the occlusal surfaces were noted, thus leading to failure ofthe automated caries scoring algorithms. This i expected to be observed ina clinical setup a8 wel, and the operators should be aware of such limitations. Despite the good performance of IOS algorithms for caries detection and classification, thee is still possibility for future algorithm improvement and implementation of other parameters, such asthe surface arca of caries lesions, in order to improve diagnostic performance. Incorporating the lesion surface area in the algorithms can potentially prevent the fase classification of some narrow initial arrested cares lesions as mote extensive lesions ‘duc to dark stains, There may also be potential inthe assessment of caries lesion activity using this system by cxamining red fluorescence from the dental plaque, estimating lesion size change over time (Le. monitoring). and obtaining information on surface roughness" all worth investigating ™- Lastly, the development of advanced algorithms based on machine learning seems promising given the recent advances in this field”! Scientific Reports) (onsyansn76 | beeps idl org/L01036)541598-021.00259.». natureportfoliowww.nature.com/scientificreports! Conclusion “The automated algorithms for occlusal caries detection and classification accompanying the IOS system were validated against histology shoving an overall comparable in vivo diagnostic pertormance tothe visual exami nation. The algorithms can be used both for in vito and in vivo assessments, Only minor differences between their in vito and in vivo diagnostic performance were noted, “This novel system exhibits encouraging performance for clinical application on occlusal caries detection and classification, while diferent approaches canbe investigated for potential optimization of the system. Received 4 Fane 2021; Acoopted: 24 September 2021 Published online: 8 October 2021 References 1. Mich. a Development of horescence-hased as scoring system for an inaoral scanner: Amn vito sty. Cris Res '34 324355 2020 hp args com! Arle/PalTexts09925 tp wenn oh go pubmed 053052 2. Micha, 5. Vane. Hstrand, Ka Benet, A. Detecting carly rset wea aig a snraoral scanner system [Dn 100, 10844, ps doar 10.1016 ent 2020103445 (203) 4 Zhang, Huang, Z, Ca Lana, Digital asessment of gingiva morphological changes and elated ators fer inl ps ‘ont therapy fOr 8, 9-64 (2021) 4 Chen, fi. Ky Zh H, Sli HS 8 Wes KD dnsbuton of deat plague on oslo sce ing 2D-aoescense- nage {9D suriceelstation, Comput. Bil Med. 123, 103860 (2020), ips linkinghubsevet comet pS00101525209021, Estero ¥ Hse Lasers nen wih 2-H tannin ath a9 Danis ht sy spo Sere rt. a el Sse 2 03) 4 Dayoan ca cont Opt dtc ets ortereersnaon sey cs ‘tse i t2-88 13 wena lhcen/ jr 7. Ea Man Aanagunga, Rages Tan gos crc tetra the dco f al fs Asuna Bae art tea ye hog toe) &. Fog B Gy minting demos snc Wier Sah. #8 Asn te) Ou Digs "apne using 0) 9. Sone ek Re Ailsa Chne an henge es 9879-74) {pmlc ape pacts nec CN un Senko theoany. Pa bruce bk} Dep amin cares detest ened igamiioston nag plot} Dna ole 0 (0) 1. FSR mies A Alpha del ae den wigan tp eal nto ariestre Faye hp tongs ay porns EN v2 frost etal comps eed wey fe dees nd dtcton f oh cis om Photog nage: Cpu Blo 9-5 pte} amponed S056 Ol) va, Huet cat tela Cas Dan a nen Sr 1CDAS A Cra a 82108-18918) we Enercon 14, ey Cae detection gos Noell Dt 4727-79 (2ME) ip ing cv com ‘ipso ote 15 Symington gunn hangin cml rie wl oo ans 25°5 (ab) ore leercon aes l 30 1 Aemcc& Raing Eetfnigg whlsance enhance chy rly se Ses in dona ai 17, Xn sal aon Pee ng wh Rela shame IRE) fr ing eam deacon Invi Cn a, 231-299 G01 were 18 Ando sere Sky 6 K-87. tot nai Gero tig Natl Whi Spt Less Ug ‘tative ighinael rence: Cn 38,39 ie omc eae 1s nl Nm sat Grune 8 Key Mao terme suo rc bs camer frdesibn fata ens Clin On beige 1 20, Minch abso eVects) Okne, FS lina ation fi rbiogaphis nde hess meod sn sedi i 13-1208, Vamenta faeces ain foil caving 95-61 Cnt pti composing crs po soasatane OT a1 22. Dun hea The proman of escona an freee Bd ch rac Cars ett: A vo ‘yh tle ian eon Ana ah 950 D2 heer sauna 013017 23, Ridder MF Sail mcd org tc rede of dace henge ta eatin or ty Selo ad ny M3 98-0 fri pss eke 0). 24, Bt Real Deaton acy ast pm otal aso Ame sly. Ope: Det. 32, BEEISan errr penned se eh yeeros 25, ma sft entra Cas Delon sn Aneesh (CDAD anita en etn Seal ‘Ste Conan Det Ora co 5 1951 pe scp OAT 1 aS SEs OM ae. Fa aro Kirsten ascent (CN) an ner ce ction enumerate (CER so sagt aes ee smi a Cooma Bit Ort ipo ts ) 27, Pay A gr W gh, $M Tet abi gon QLP es. OR 2928-73. pei. Syn ee aera tan 28, Delong Delong Cle en DL. Conga the acs dr or cnc er opting ante crve haptic ppc Bomar, (3). 2». Rims hin Dwr beset ss spn fm gre Sa hd Med Rs dein on) 20 lon Samar tl mpac of ring gl or mile clio eine fg cache al {CNet Fn 09h Von Naa ta pact of en po ae cl non eis aga cnc aog free tds Le ed 2 3-28 OND Scientific Reports|_ (202:)1.28276 | beeps idl org/L01036)541598-021.00259.». nature portfoliowww.nature.com/scientificreports! 22. Gender TS. Mato for tig seo and psy fds a A Taal Ra» Radi 48 (01 pig nono ape aad 20) DC os. Faces, at inno Dire Sore Mca on Ln exec alc: Two Stay eer 4, 81185 (200, wr ager sonra ce ale 3238 M4. Quint M Avena De Alea, Ait Hoke Yas stein tee ane anier iy tts cae gree ean na) Pa He ow 6,9) la 25, Kn ling sd ating i bi ice agp Jl Ma 38,23 202, 3 Han 1 Ki Ko Hy os Hh tL Ascingte of unten ene il a Sinaia cst lag Paty Tei, 3429 pal ep 2132.00 G0) Stem eat Re rect tye thant erg) nar 1-9 D6) Ande Sits: Dacminaton cee leion cy: Releon nd on or harcaaton fee foeeston Oper Dee 1301-30 01 a». RoEse cl Acton cece quate mens ied sing ose ed mend ait ae ccna par lr af Prats Be 2-2 ‘Acknowledgements ‘The authors acknowledge the laboratory technician Lisette Laren for her assistance with the sample storage and preparation, the development teams at 3shape A/S for technical support, and Innovation Fund Denmark for financial suppor. Author contributions ‘SM. contributed tothe study design, conducted the analyses using IOS algorithms and histology, and drafted the ‘manuscript. MS.L. defined the algorithm ALG4 and implemented all the algorithms in the prototype software used in this study: PN. collected the teeth, conducted the visual examination, and scanned the teeth using the 10S, A.B, coordinated the laboratory activities. A.R.B, A.B, KRLE., CR, and CV. contributed in the study design. ll authors reviewed and critically revised the manuscript Competing interests ‘The current study was funded by Innovation Fund Denmark (Grant no, 8053-00005B). Based on the founda- tion’ guidelines and according t an industrial PhD agreement between the industrial partner 3Shape TRIOS, A/S and the University of Copenhagen, SM. salary is patially covered by 3shape TRIOS AIS, Furthermore, the co-authors C-V-and MSL, are employed at 3shape TRIOS A/S. The other co-authors A.R.B. A.B, CR, PN and KRE. declare no conflicts of interest. Additional information Supplementary Information The online version contains supplementary material avaiable a btips//doi.org/ 10.1038 /+41598-021-00259-v Correspondence and requests for materials should be addressed to SM Reprints and permissions information i available at www natuee com/seprints Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional afiliations. GIDE] Open Access This ete lcnsed under Creve Commons Anibution 4.0 atermtional Likene which pernits se sharing adaptation, dtbution ad eprodcton in a nem or format ar long at Fou give sppoprnteceit othe orignal author) ap he soars, provide sik othe vc Comma cence an nate changes were nae. The ingest hed pry material in his sre ae incded in the arcs Cresive Commons cence, less indeed otherwise a cet ie the materi If tera i ot inched inthe ais Crenve Commons lence snd your itended tse ft ermine by antoryeplation ov excels the permite we you wl ped to cain permission ety fm {Recoprigh olde To ew copy of hs coce, vist tp cetecommonsenpliceneby 4.0 © The Author(s) 2021, corrected publication 2022 Scientific Reports) (onsyansn76 | beeps idl org/L01036)541598-021.00259.». natureportfolio