EditorialDOI: 10.3290/j.ijcd.b4702447, PubMed ID (PMID): 38014639Pages 279-280, Language: English, GermanBeuer, Florian
ScienceDOI: 10.3290/j.ijcd.b3840521, PubMed ID (PMID): 36705319Pages 285-299, Language: English, GermanTop, Ahmet Esad / Özdoğan, M. Sertaç / Yeniad, Mustafa
Ziel: Während der Einsatz von Deep-Learning-Modellen auf verschiedenen Gebieten in einer Vielzahl von Studien untersucht wird, existieren nur wenige solcher Studien zur Bildgebung in der Zahnmedizin. Ziel des vorliegenden Beitrags war eine Untersuchung der Leistungsfähigkeit faltender neuronaler Netze (Convolutional Neural Network, CNN) bei der Erkennung und Bestimmung von Quantitätsniveaus dentaler Restaurationen (Kronen und Brücken) auf Panoramaschichtaufnahmen (PSA) unter Verwendung eines neu zusammengestellten Datensatzes.
Material und Methode: Insgesamt wurden 20 973 PSA verwendet, die von drei Spezialisten nach fünf Kategorien (0 Restaurationen [„keine Restauration“], 1–3 Restaurationen [„geringes Restaurationsniveau“], 4–6 Restaurationen [„mittleres Restaurationsniveau“], 7–11 Restaurationen [„hohes Restaurationsniveau“], 12–32 Restaurationen [„sehr hohes Restaurationsniveau“]) gelabelt wurden. Auf diesem Datensatz wurden AlexNet-, VGG-16- sowie mehrere ResNet-Varianten- Modelle trainiert und bezüglich ihrer Leistungsfähigkeit bei der Klassifizierungsaufgabe bewertet. Für alle Versuche wurde eine 10-fache Kreuzvalidierung (d. h. jeweils 9 Teile als Trainingsdatensätze und ein Teil für die Validierung) und eine Datenaugmentation durchgeführt.
Ergebnisse: Die besten Ergebnisse mit einer Genauigkeit von 92,7 % lieferte das ResNet-101-Modell. Auch der Makro- Durchschnitt der Fläche unter der ROC-Kurve (Area under the Curve, AUC) war für dieses Modell mit 0,989 am größten. Die Genauigkeiten der anderen Modelle für den Datensatz waren: AlexNet = 75,5 %, VGG-16 = 85,0 %, ResNet-18 = 92,1 %, ResNet-50 = 91,7 % und InceptionResNet-v2 = 92,1 %.
Schlussfolgerungen: Eine Genauigkeit von 92,7 % ist für ein computergestütztes Diagnosesystem überaus vielversprechend. Das Ergebnis beweist, dass ein entsprechendes System dem Zahnarzt assistieren kann, indem es unterstützende Vorab-Informationen ab dem Zeitpunkt der ersten Panoramaschichtaufnahme des Patienten liefert. Da der hier eingeführte Datensatz umfassend genug ist, kann er für andere Problemstellungen umgelabelt und in weiteren Studien verwendet werden.
Keywords: künstliche Intelligenz, computergestützte Diagnostik, convolutional neural networks, CNN, faltendes neuronales Netz, Deep Learning, Restauration, Panoramaröntgen, Panoramaschichtaufnahme
ScienceDOI: 10.3290/j.ijcd.b3840535, PubMed ID (PMID): 36705317Pages 301-309, Language: English, GermanAdnan, Niha / Khalid, Waleed Bin / Umer, Fahad
Aim: To develop a deep learning (DL) artificial intelligence (AI) model for instance segmentation and tooth numbering on orthopantomograms (OPGs).
Materials and methods: Forty OPGs were manually annotated to lay down the ground truth for training two convolutional neural networks (CNNs): U-net and Faster RCNN. These algorithms were concurrently trained and validated on a dataset of 1280 teeth (40 OPGs) each. The U-net algorithm was trained on OPGs specifically annotated with polygons to label all 32 teeth via instance segmentation, allowing each tooth to be denoted as a separate entity from the surrounding structures. Simultaneously, teeth were also numbered according to the Fédération Dentaire Internationale (FDI) numbering system, using bounding boxes to train Faster RCNN. Consequently, both trained CNNs were combined to develop an AI model capable of segmenting and numbering all teeth on an OPG.
Results: The performance of the U-net algorithm was determined using various performance metrics including precision = 88.8%, accuracy = 88.2%, recall = 87.3%, F-1 score = 88%, dice index = 92.3%, and Intersection over Union (IoU) = 86.3%. The performance metrics of the Faster RCNN algorithm were determined using overlap accuracy = 30.2 bounding boxes (out of a possible of 32 boxes) and classifier accuracy of labels = 93.8%.
Conclusions: The instance segmentation and tooth numbering results of our trained AI model were close to the ground truth, indicating a promising future for their incorporation into clinical dental practice. The ability of an AI model to automatically identify teeth on OPGs will aid dentists with diagnosis and treatment planning, thus increasing efficiency.
Keywords: artificial intelligence, deep learning, dentistry, neural networks, convolutional neural network, intraoral radiography
ScienceDOI: 10.3290/j.ijcd.b3840393, PubMed ID (PMID): 36749284Pages 311-317, Language: English, GermanPark, Jun Hyub / Lee, Du-Hyeong
Aim: The present study aimed to evaluate the accuracy of automated detection of preparation finish lines in teeth with defective margins.
Materials and methods: An extracted first molar was prepared for a full veneer crown, and marginal defects were created and scanned (discontinuity of finish line: 0.5, 1.0, and 1.5 mm; additional line angle: connected, partially connected, and disconnected). Six virtual defect models were entered into CAD software and the preparation finish line was designated by 20 clinicians (CAD-experienced group: n = 10; CAD-inexperienced group: n = 10) using the automated finish line detection method. The accuracy of automatic detection was evaluated by calculating the 3D deviation of the registered finish line. The Kruskal-Wallis and Mann-Whitney U tests were used for between-group comparisons (α = 0.05).
Results: The deviation values of the registered finish lines were significantly different according to conditions with different amounts of finish line discontinuity (P < 0.001). There was no statistical difference in the deviation of the registered finish line between models with additional line angles around the margin. Moreover, no statistical difference was found in the results between CAD-experienced and CAD-inexperienced operators.
Conclusions: The accuracy of automated finish line detection for tooth preparation can differ when the finish line is discontinuous. The presence of an additional line angle around the preparation margin and prior experience in dental CAD software do not affect the accuracy of automated finish line detection.
Keywords: tooth preparation, marginal defect, computer-aided design, finish line, automated detection, computer algorithm
ScienceDOI: 10.3290/j.ijcd.b3839037, PubMed ID (PMID): 36749283Pages 319-330, Language: English, GermanXu, Shu-Xi / Tong, Xue-Lu / Tan, Fa-Bing / Yu, Na / Ma, Chao-Yi
Aim: The aim of the present study was to evaluate the effect of cement gap and drill offset on the marginal and internal fit discrepancies of crowns designed with different tooth preparations.
Materials and methods: Five tooth preparations were constructed, and crowns with different cement gaps and drill offsets were obtained. Then, best-fit alignment was performed on the crowns with the corresponding tooth preparations, and the fit discrepancies were expressed by color-coded difference images and root mean square (RMS) values. The RMS values of each group were analyzed by the rank-based Scheirer-Ray-Hare test (α = 0.05).
Results: The color segments in the sharp line angles area of the Sharp line angles group changed significantly before and after the drill offset. The cement gap had a significant effect on the marginal, internal, or overall fit discrepancies of the five design groups (P < 0.001), while the drill offset had a significant effect on the marginal fit discrepancies of the Shoulder-lip group and the internal or overall fit discrepancies of the Sharp line angles group (P < 0.001). Additionally, the interaction effect between cement gap and drill offset was significant for the marginal fit discrepancies of the Shoulder-lip group and the internal or overall fit discrepancies of the Sharp line angles group (P < 0.01).
Conclusions: The cement gap and drill offset had a significant adverse effect on the marginal or internal fit discrepancies of the crowns designed with the shoulder-lip and sharp line angles designs. Tooth preparation designs with intense curvature changes such as shoulder-lip and sharp line angles should be avoided clinically.
Keywords: computer-aided design, drill offset, cement gap, tooth preparation design, fixed crown, fit discrepancies
ScienceDOI: 10.3290/j.ijcd.b3839017, PubMed ID (PMID): 36749282Pages 331-337, Language: English, GermanLi, Rong / Zhang, Rui / Zhou, Yongsheng / Peng, Juanhong
Aim: The aim of the present in vitro study was to assess and compare the accuracy of two best-fit alignment strategies with different reference areas for wear measurement with an intraoral scanner (IOS).
Materials and methods: Eight anatomic contour zirconia crowns were fabricated and scanned twice with an IOS. One of the scan datasets (Data Trueness) was duplicated and wear facets were simulated (Data Wear). The other scan dataset (Data Baseline) was aligned to Data Wear by two best-fit alignment strategies with different reference areas (the occlusal surface with no signs of wear [Group Occlusal], and the axial surface [Group Axial]), and 3D deviation analysis was performed to detect wear loss. The 3D deviation between Data Trueness and Data Wear was calculated as the truth-value for accuracy evaluation (Group Trueness).
Results: The color-difference map showed Group Occlusal had a similar wear-facet distribution to Group Trueness while Group Axial showed an obvious tilting position, and the obtained height loss values were larger and with large standard deviations. Both Group Occlusal and Group Axial showed significant differences compared with Group Trueness in maximum height loss and mean height loss (P < 0.05) while showed no significant difference in mean distance (P > 0.05). The paired t test showed significant differences between Group Occlusal and Group Axial in maximum height loss and mean height loss (P < 0.05) while showed no significant difference in mean distance (P > 0.05).
Conclusions: Best-fit alignment with the occlusal reference area produced a better alignment result than that with the axial reference area. Measuring wear with an IOS has potential, but the method is prone to overestimating the height loss.
Keywords: wear measurement, intraoral scanner, best-fit alignment, accuracy, digital, in vitro
ApplicationDOI: 10.3290/j.ijcd.b4653531, PubMed ID (PMID): 38014640Pages 339-346, Language: English, GermanCamps-Font, Octavi / Vilarrasa, Javi
Aim: To present a minimally invasive approach to expose palatally displaced canines (PDCs) using a surgical guide.
Materials and methods: Surgical guides for palatal canine exposure are fabricated with CAD/CAM technology. With adequate software, it is possible to match the STL files of the dental arch with the DICOM images of the maxilla. On the STL 3D model file, the operator can localize and determine the exact position of the impacted canine. In turn, this allows the identification of the ideal location of the window. A software application facilitates the design of the surgical guide, which is printed using a 3D printer.
Results: Exposure of PDCs can be achieved satisfactorily using surgical guides.
Conclusions: The use of computer-guided surgical exposure of PDCs allows both the reduction of surgical time and surgical invasiveness, minimizing patients’ postoperative discomfort. Controlled clinical trials are necessary to evaluate more fully any advantages of this minimally invasive technique.
Keywords: canine impaction, canine exposure, CAD/CAM, oral surgery, orthodontics, surgical guide
ApplicationDOI: 10.3290/j.ijcd.b3960939, PubMed ID (PMID): 36928755Pages 347-363, Language: English, GermanGoob, Janosch / Prandtner, Otto / Schweiger, Josef / Güth, Jan-Frederik / Edelhoff, Daniel
Pronounced defects of the dental hard tissue can be caused by different etiologic factors. Most frequently, they are associated with changes in the vertical dimension of occlusion (VDO), which may also influence the condylar positions. These defects can lead to irreversible loss of tooth structure and have dramatic functional and esthetic consequences, often requiring complex rehabilitation. In this situation, CAD/CAM-fabricated occlusal splints made of tooth-colored polycarbonate are a proven and safe pretreatment approach in terms of esthetics and function. Rebuilding lost dental hard tissue to restore the occlusion and VDO to an adequate condylar position is a prerequisite for any sustainable and functional rehabilitation. In the future, digital systems will support this complex process, customizing it and making it simpler and more precise. The DMD-System (Ignident) provides patient-specific jaw movement data to optimize the CAD/CAM workflow. This system allows real movement patterns to be digitized and analyzed for functional and potential therapeutic purposes, integrating them into the dental and laboratory workflow. In the present case, the familiar tooth-colored CAD/CAM-fabricated occlusal splint is supplemented by digital centric jaw relation recording and individual movement data.
Keywords: vertical dimension of occlusion (VDO), instrumental functional analysis, maximum intercuspation (MI), maximal intercuspal position (MIP), centric condylar position (CCP), centric relation (CR), tooth-colored occlusal splint, digital workflow
PubMed ID (PMID): 38014641Pages 365-368, Language: GermanKordaß, Bernd / Schlenz, Maximiliane
Online OnlyPubMed ID (PMID): 38014638Pages 1-4, Language: EnglishKordaß, Bernd / Schlenz, Maximiliane