Accuracy and Time Efficiency of AI Versus Manual Segmentation in Adult Crowding Using Aligner Software

This study is a prospective diagnostic accuracy study designed to evaluate the accuracy and time efficiency of artificial intelligence (AI)-based automated tooth segmentation compared with manual segmentation in adult patients with moderate to severe dental crowding.

Digital orthodontic workflows depend on accurate identification and separation of individual teeth from three-dimensional dental models. Accurate tooth segmentation is important for diagnosis, assessment of dental crowding, virtual treatment planning, clear aligner design, and evaluation of planned tooth movements. Manual segmentation is currently used as a reference method because experienced orthodontists can identify and delineate individual tooth boundaries accurately. However, manual segmentation can be time-consuming and may be affected by differences between operators, particularly in crowded dentitions where overlapping teeth, proximal contacts, unclear boundaries, and gingival margins can make tooth separation more difficult.

Recent advances in artificial intelligence have resulted in automated tooth segmentation tools that can identify and separate individual teeth from digital dental models. These systems may reduce the time required for digital orthodontic workflows; however, their performance may be affected by the complexity of the dentition. In particular, dental crowding may create areas in which tooth boundaries are difficult to distinguish, potentially resulting in deficient or excessive segmentation or failure to correctly identify individual teeth.

The present study will evaluate AI-based segmentation using two orthodontic software platforms, Archform 2.6.4.1 and Blue Sky Plan V5. Pretreatment intraoral digital scans of adult patients with dental crowding will be used. The digital scans will be obtained using a Medit i900 intraoral scanner and will be available as digital dental models in STL format. The digital models will be anonymized and coded before analysis.

For the index test, the STL models will be imported into Archform 2.6.4.1 and Blue Sky Plan V5. The AI-based segmentation functions of each software platform will be used to automatically identify and separate individual teeth. The resulting segmented models will then be evaluated for successful identification and separation of the teeth and for the quality of the generated tooth boundaries.

Manual segmentation will be performed on the same digital dental models by experienced orthodontists and will serve as the reference standard. The orthodontists will delineate the boundaries of individual teeth using the available digital segmentation tools. The manual segmentation procedure will be performed independently from the AI-generated results to minimize the possibility of influence between the index test and the reference standard.

The study will evaluate segmentation performance at the individual tooth level. A segmentation will be considered successful when the software correctly identifies the tooth and accurately delineates its margins. Segmentation will be considered unsuccessful when tooth margins are deficient or excessive, when the boundaries are incorrectly identified, or when an individual tooth cannot be correctly identified or separated.

The accuracy of AI-based segmentation will be assessed by comparing the AI-generated tooth segments with the manually segmented reference models. Mesiodistal tooth width measurements will be obtained from both AI-generated and manually segmented models. Differences between the measurements will be evaluated to identify discrepancies associated with automated segmentation. The study will also document specific segmentation errors, including deficient or excessive segmentation, tooth identification failures, proximal contact reconstruction problems, and difficulties in detecting gingival margins. Errors involving occlusal, vestibular, and lingual surfaces will also be recorded.

The study will additionally assess the time efficiency of the three segmentation approaches: AI-based segmentation using Archform 2.6.4.1, AI-based segmentation using Blue Sky Plan V5, and manual segmentation. Processing time will be recorded from the beginning of STL file import or segmentation processing until completion of the segmentation procedure. The mean time required by each method will then be compared.

To reduce assessment bias, the digital models used for evaluation will be anonymized and coded. Examiners assessing segmentation success or failure will be blinded to whether the models were produced using AI-based or manual segmentation. The order of model assessment will be randomized.

The study will focus specifically on adult dentitions with moderate to severe dental crowding, as crowding presents a clinically challenging environment for automated tooth segmentation. The study will therefore provide information not only on the overall performance of AI segmentation but also on the types of errors that may occur in crowded dentitions.

The primary purpose of the study is to evaluate the success and failure of AI-based tooth segmentation using Archform 2.6.4.1 and Blue Sky Plan V5 compared with manual segmentation. Secondary purposes include comparing the performance of the two AI-based software platforms, evaluating differences in processing time, and identifying the types of segmentation errors associated with adult dental crowding.

The results of this study may provide evidence regarding the reliability and efficiency of AI-assisted tooth segmentation in challenging orthodontic digital workflows. The findings may help determine the extent to which automated segmentation can support orthodontic diagnosis and digital treatment planning and whether manual review or correction may remain necessary when working with crowded dentitions.