From pixels to precision: a narrative review of AI-driven 3D morphological analysis and digital twinning in alveolar cleft management

Introduction

Cleft lip and palate (CLP) is a common craniofacial anomaly that impacts oral and maxillofacial function, appearance, psychological health, and social life (1). Alveolar bone defects are commonly seen in CLP patients and should be carefully considered for treatment planning (2). To enhance the accuracy of alveolar defect evaluation, artificial intelligence (AI) and three-dimensional (3D) imaging can be used to tailor treatment strategies for CLP patients (3).

The use of 3D analysis of alveolar bone defects is highly beneficial, especially with the help of AI and deep learning techniques. Convolutional neural networks (CNNs) and other models can be used to automatically detect, segment, and classify these defects, thus aiding in quantitative assessment and treatment planning. For example, the accuracy is improved by using a multi-view 3D surface model with automatic severity classification algorithms. Furthermore, algorithms can be made explainable, so they can be used in clinical applications with high reliability and transparency (4).

The 2D radiography was the only tool used traditionally for clinical assessment and was not as useful when assessing more complex volumetric structures, such as alveolar clefts. This has changed with the introduction of cone-beam computed tomography (CBCT)—the technology now provides high-resolution volumetric data. AI-based CBCT image analysis has proven to be effective in the diagnosis of periodontal bone loss and the morphology of bone defects and has been applied for the early and accurate diagnosis of alveolar bone defects (5,6).

Alveolar bone defects are related to genetic and environmental factors that affect oral function and facial esthetics in CLP patients. However, despite its drawbacks, bone grafting is the most common treatment. The potential of 3D imaging and AI for accurate diagnosis and personalized treatment is promising. This paper summarizes the problems and prospects of intelligent sensor data, clinical decision-making, and AI for better clinical results in this area (Figure 1). We present this article in accordance with the Narrative Review reporting checklist (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0268/rc).

Figure 1 From pixels to precision: AI-driven 3D morphological analysis and digital twinning in alveolar cleft management. AI, artificial intelligence; CBCT, cone-beam computed tomography; CNN, convolutional neural network.

Methods

A narrative review was conducted to investigate the latest uses of AI in the diagnosis, treatment, and post-treatment care of alveolar clefts. The approach employed in the literature search is briefly outlined in Table 1. The studies published from January 1, 2020, to December 31, 2025 were identified and analyzed, and the function and clinical applications of AI in alveolar cleft-related research and management were summarized. A literature search was carried out via the PubMed database. The keywords used were: “Artificial Intelligence”, “Alveolar Cleft”, “Cleft Lip and Palate”, “Digital Twin”, and “Cone-Beam Computed Tomography”.

Data cornerstone: intelligent sensing and quantification of alveolar cleft defects

“Seeing” defect: evolution of 3D automatic segmentation

With the developments in the last few years in medical imaging technologies, 3D imaging is now more important than ever in the diagnosis and treatment planning of congenital alveolar clefts. One of the key steps is lesion visualization, and 3D automatic segmentation has replaced conventional 2D segmentation algorithms in more advanced deep learning-based algorithms, resulting in improved segmentation accuracy and efficiency.

However, traditional approaches, such as region-growing thresholding, can only achieve rapid segmentation with limited image quality and complexity of anatomical structure (7). Since then, CNN-based architectures such as 3D U-Net and V-Net have emerged as the leading methods for handling spatial features, being able to do so more efficiently (8). Segmentation based on AI has shown good results in assessing alveolar cleft bone grafting and tooth auto-segmentation, proving more efficient and precise than manual segmentation (9,10) and leading to a significant reduction in segmentation time with high-Dice-coefficient models (11,12). For uncommon but serious errors and in complex or critical regions, however, physician verification is still required. This verification burden can be reduced in the future with the integration of uncertainty quantification with clinically guided evaluation metrics (13,14).

However, there are still some limitations. Alveolar defects are often accompanied by low-dose noise and artifacts in CBCT images, which tend to degrade segmentation performance (15), in addition to the low inherent contrast of CBCT imaging (16), reliance on local image features, and the relatively low robustness of the model (17). It is, therefore, difficult to discriminate microstructural details, such as the boundaries of bone defects, leading to segmentation margins that are inaccurate or discontinuous. A lack of accurate and annotated 3D data makes model generalizability and effective training more difficult (18).

A number of optimization techniques have been proposed to overcome these limitations. Attention mechanisms can be added to segmentation networks to focus the network’s attention on relevant anatomical structures, thereby decreasing the effect of irrelevant information and improving the segmentation of these structures with more accuracy and robustness (19). Transfer learning, which uses a pre-trained model to solve a similar problem (20), can reduce difficulties in dealing with the challenges of small-sample studies. Moreover, data augmentation approaches, such as the addition of noise, rotation, and/or scaling images, increase the diversity of training data and improve the model’s generalizability (21). Combined, these techniques may facilitate the application of more complex treatments in a wider context, using 3D auto-segmentation of these types.

The 3D auto-segmentation tool, based on deep learning, has greatly improved the assessment of alveolar clefts and other complex structures. Advanced techniques like attention mechanisms, transfer learning, and data augmentation are being used to further improve segmentation accuracy, aiding in the diagnosis and treatment planning of congenital alveolar clefts.

“Quantitation” of defects: multidimensional morphological analysis beyond volume

The key method used for the evaluation of congenital alveolar cleft has been the volume measurement of the defect. After the integration of 3D imaging and AI, quantitative evaluation has now expanded to multidimensional morphological evaluation (4). Current techniques evaluate bone defects according to defect height, width, complexity, and morphology to obtain a more holistic picture of the bone defect and tailor treatment (22).

Spatial topology and/or geometric heterogeneity for accurate surgical planning is not described by volumetric assessment, the standard method. Clinically relevant parameters such as defect height and width, and surface irregularity, were evaluated using AI-based morphometric analysis, as surface morphology is directly related to the graft adaptation and stability of fixation. AI can be used to perform a more detailed anatomical risk profile of adjacent structures, such as the nasal floor and the pyriform aperture, in addition to topological analysis of the volume of this region (23).

Evaluation of the adjacent anatomic structures is also important and should be quantified. The severity of the defects and the technical complexity of the surgery are well correlated with the changes involved in the roots of teeth, the floor of the nose, the pyriform aperture, and the palate (24,25). The eruption status and spatial location of alveolar defects and margins are factors associated with bone grafting success (26); the changes of the nasal floor and pyriform aperture are associated with airway patency and protection of neurovascular structures (27); and the morphology of the palate is associated with oral functional recovery after surgery (28). Automated analysis of these complex structures may improve diagnostic accuracy and reduce clinical workload (4).

The key benefit of multidimensional morphometric analysis is its ability to adequately describe the complex features of an alveolar defect, rather than relying only on volumetric approaches. The AI analysis of multiple morphological parameters offers more detailed information to assist with surgical planning and the assessment of prognosis. Explainable AI techniques can enhance the trust a clinician has in the algorithm’s decisions (29), and multidimensional analysis can assist in the design of grafts, ensure accurate customization and intraoperative guidance, thus improving therapeutic effectiveness and enhancing patient quality of life (4).

“Understanding” defect: multidimensional characteristic-based intelligent phenotype

The traditional grading systems of congenital alveolar clefts, for example, the Bergland scale, rely on morphology and overlook genetic, tissue, and patient factors (30). This makes prediction of functional recovery, appearance, and complications less individualized and less reliable.

The image morphological properties, the parameters of bone regeneration, and the clinical features were combined as a multidimensional data space to classify alveolar cleft defects for automatic recognition of complex bone defects using machine-learning algorithms. Explainable AI has proven to be effective for the classification of alveolar defects according to shape, height, and width (4), and another algorithm combining rigid image registration and AI-based automatic segmentation has been found to be more efficient and precise than manual measurement (9).

However, there are still limitations; for example, complex defect morphologies require manual refinement following automatic segmentation, and there is morphological variability across patients that requires larger sample sizes. Discrepancies in the technique of the assessor may also affect the consistency of postoperative bone grafting evaluation.

With advances in AI and machine learning, time-sequence data-based bone defect variation prediction models have become an emerging research focus. Using 3D image data and clinical markers collected during patient follow-up, machine-learning models can support dynamic, personalized bone defect progression prediction and provide a scientific basis for clinical intervention. Combining sequential CT or CBCT data with bone defect volume and density allows modeling of bone regeneration and resorption. Finite element analysis (FEA) has been widely applied to predict mechanical stress distribution in bone defects, elucidating how the biomechanical environment affects bone regeneration (31,32). Multidimensional mechanobiology-based models coupled with machine learning further simulate cell migration, tissue formation, and other key repair processes, yielding more accurate healing predictions (33,34). Introducing FEA and mechanobiology-based machine learning into alveolar cleft bone grafting research could, therefore, simulate graft healing and support continuous surgical improvement.

AI has advanced bone grafting and repair through preoperative prediction models, deep-learning image analysis, and 3D printing, all of which have improved surgical outcomes in alveolar cleft treatment (4). In the future, AI holds significant potential to refine risk prediction, treatment design, and postoperative evaluation, advancing precision medicine in congenital alveolar cleft care.

Clinical translation: smart decision-making from pixels to patient prognosis

Assisted surgery planning: AI-empowered “preoperative sandbox”

The key to a successful surgical outcome and to minimizing surgical complications in the management of congenital alveolar clefts is careful preoperative planning (35). With the application of AI and 3D image analysis, a highly personalized “preoperative sandbox” for protocol design in surgery allows the development of a more scientifically rational surgical protocol and more efficient, precise clinical decisions (36).

Personalized bone grafting protocol design: AI-assisted autologous bone fragment/bone substitute simulated matching and design

In the treatment of alveolar defects, personalized bone grafting depends on the surgeon’s subjective choice of the bone material used. With the advent of AI and 3D imaging, it is now possible to precisely segment the bone defects in a CT or CBCT scan, and deep learning algorithms can be used to create accurate 3D bone models (15). These systems have the ability to characterize defect morphology, size, and spatial relationships and to match them to autologous bone or custom-made substitutes to best fit the anatomy.

A study using AI-assisted 3D reconstruction of teeth, alveolar bone, and the maxillary sinus demonstrated high anatomical accuracy and agreement with surgical planning, suggesting the safety and reliability of this method (37). The strategies used for alveolar cleft bone grafting are similar, allowing the surgeon to refine the protocol for bone fragment design, optimize fragment shape, and minimize the time and risk required to adjust the fragments (38). When combined with 3D printing, it has helped to manufacture precisely customized bone substitutes, thus offering a model for the intraoperative use of experimental bone substitutes (39).

In the estimation of graft volume, a multicenter study has developed and tested a deep learning model with a 3D U-Net architecture for automatic segmentation of the unilateral alveolar cleft region. In a test set of 33 patients, the Dice similarity coefficient was 0.78, while the manual to automatic segmentation consistency obtained was acceptable in 82–94% of cases. Manual segmentation took 6.5 to 14 min, whereas automatic segmentation took only a few seconds (40). State-of-the-art deep learning models, such as SegResNet, have achieved high precision in bone structure segmentation from dental CBCT images, comparable to semi-manual methods (15). AI-driven bone grafting protocol design enhances graft matching accuracy, streamlines the surgical procedure, and reduces complications.

Clinical decision support system: smart recommendation of surgery timing and approach based on AI quantitation characteristics

Surgery timing and selection of the appropriate surgical procedure are important factors that affect the treatment outcome of patients with complex congenital alveolar clefts. Traditional clinical experience-based and radiological assessments are subject to bias and inherent limitations. By applying deep learning to large, multidimensional clinical datasets and 3D features, AI enables the development of smart decision-support systems capable of recommending surgery timing and approach.

The traditional approach to surgery timing is based on clinical experience and imaging evaluation, usually during the mixed dentition phase. AI can be used to make more objective decision-making by accurately segmenting the alveolar ridge fissure defect and estimating bone-graft volume from the CBCT images (40). This preoperative simulation helps the surgeon determine bone defects and the required bone volume, which are essential for scheduling the appropriate surgical time and planning successful bone transplantation to maintain dental arch continuity and the permanent eruption of teeth (41). Age, growth and development status, and dentition stage are used by AI-based systems to undertake quantitative analysis of bone, soft tissue, and related anatomical structures, which not only helps to develop optimal surgical regimes, but also helps to assess surgical risk and predicted outcomes (42). For congenital alveolar cleft, 3D image analysis with AI has been demonstrated to help narrow down surgical approach selection and increase the accuracy of the analysis (43).

In terms of planning, some studies have focused on the occurrence of palatal fistulas during secondary alveolar bone grafting without AI involvement, and this has implications for the management of fistulas by AI. Early fistula repair is a strong risk factor for late fistula (OR = 17.17), and even up to 43% of fistulas can recur (44). Incorporating these clinical factors, such as fissure type and early repair history, into future AI models could aid in predicting individual fistula risks and optimizing surgery timing. The available predictive models for alveolar cleft surgery timing rely mostly on basic anatomical, radiological, and demographic data. While socioeconomic factors and psychological expectations are known to be clinically important (45,46), they are not commonly included as input features, perhaps because socioeconomic factors are difficult to quantify (47). This is an important reason why future model development is needed.

Treatment response prediction: moving towards precise prognosis

Accurate estimation of the likelihood of treatment success is key to personalized medicine in the treatment of congenital alveolar clefts. The prediction of therapeutic response and the long-term prognosis is now becoming more and more possible using AI, multimodal data fusion, and 3D image analysis.

The most important aspect of alveolar cleft repair is bone grafting to achieve stability and functional recovery of the maxillofacial structures. Until now, the traditional method of postoperative evaluation has been radiological observation and clinical follow-up, and there are no specific risk assessment methods that can be performed before surgery (48). Preoperative quantitative evaluation of the success of the bone graft and the rate of bone graft resorption is possible through the application of AI prediction models that incorporate 3D radiomics data and clinical parameters. These models remove multidimensional image features from deep learning training to predict personalized outcomes and inform both surgical and post-surgical management, highlighting different responses to bone grafts and allowing early intervention in high-risk cases.

Developmental abnormalities of the maxillofacial region include abnormalities in function and esthetics, which require bone grafting in both short-term repair of the defect and long-term growth of the maxilla (49). The dynamic evaluation of post-grafting maxillofacial development trends, the classification of bone defects at different healing stages, and the identification of bone regeneration potential and risk factors are made possible by using AI-based longitudinal analysis of multiple time-point 3D data, which provides a scientific basis for surgery timing and the selection of graft materials (50).

Orthodontic treatment is a vital component in the process of monitoring the eruption path, root position, and transplantation timing of canines. The best prognosis for natural eruption is with a secondary bone graft performed at 1/4 to 1/2 completion of canine root formation (51). If there is a dental arch collapse, then grafting before arch expansion will stabilize arch morphology. Pre-eruptive bone grafting has consistently been shown to enhance bone bridge closure and decrease resorption, and post-eruptive orthodontic gap closure has been shown to further decrease resorption (52).

While no specific application of AI to canine eruption management has been reported, future models could be developed to quantify the inclination angle of the canines, the stage of eruption, and the severity of arch collapse from CBCT scans, thereby predicting the probability of bone bridge formation and suggesting an optimal time for grafting. When these factors are incorporated into a comprehensive decision-support system, it can help minimize surgical failure and orthodontic problems and enhance functional and esthetic restorations.

Multimodal data fusion: construction of a more complete patient portrait

Multimodal data fusion integrates multiple data sources to generate a patient profile. Combining imaging data with clinical information, genomics, and proteomics can enhance the diagnostic accuracy of congenital alveolar clefts, as well as provide a better understanding of the disease pathogenesis and repair process.

Deep- and machine-learning models are able to capture complex correlations that are not reflected in either radiological data or clinical data alone, when the data is combined with other clinical parameters, including symptoms, medical history, and physiological markers, such as CBCT, MRI, and intraoral scans. Combining data types improves performance in predicting pathology and disease stage compared to single modalities (53), and 3D imaging can be complemented by patient-specific clinical information and may be useful for personalized surgical planning and dynamic disease evaluation (54).

The data from genomics and proteomics provide an alternative approach for identifying biomarkers relevant to alveolar cleft. Genetic predisposition plays an important role in its occurrence, and proteomic analysis reveals disease-related molecular mechanisms (55). Multi-omics data can be used to extract important biomarkers related to disease initiation and tissue repair using graph convolutional networks and multimodal deep learning architectures, which could reveal the molecular mechanisms underlying developmental abnormality (56,57).

To achieve deeper data fusion beyond simple data-layer superposition, effective multimodal fusion uses attention mechanisms and cross-modal learning (58). Transformer-based architectures, on the other hand, preserve long-range dependencies and complex cross-modal relationships, resulting in better generalizability and explainability (59,60). A multimodal model that incorporates 3D images, clinical data, and molecular omics data has the potential to facilitate realistic digitalization of patients for surgical planning and prognosis prediction in precision medicine for alveolar cleft (61).

Cornerstone of trust: transparency and ethical challenges of AI in clinical application

Opening the “black box”: necessity of explainable AI

The increasing use of AI in the diagnosis and treatment of congenital alveolar cleft has led to a growing need for clear clinical decision-support systems. While traditional deep-learning models have demonstrated good performance in disease diagnosis, their black-box characteristics limit the understanding of decision-making by medical practitioners and prevent their further clinical adoption (62). Clinicians need a deeper understanding of model behavior to ensure AI outputs align with clinical expertise, minimizing the risk of misdiagnosis and enhancing safety in clinical practice. This was addressed in one study, based on applying an explainable AI framework in the classification of the severity of alveolar bone defect in CLP using a multi-view three-dimensional model by applying CNN-based classification and attention mechanism to highlight regions of interest, which helps in increasing the transparency and reliability of the model (4).

Visualization tools are vital to the adoption of AI reasoning. Gradient-weighted class activation mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP) methods illustrate the most relevant regions of an image and the importance of individual features via heat maps, enabling practitioners to understand how the model makes its decisions (63). A deep-learning model was used to quantify asymmetry in one application evaluating lip symmetry after CLP surgery, in which visualization showed that the model focused on clinically relevant facial regions (64). Automatic bone graft evaluation has also incorporated AI-based postoperative bone mass measurement and explainable analysis to facilitate physicians’ interpretation of the grafting results (9).

Explainable AI actively enhances clinician trust and collaborations between patients and clinicians, enabling doctors to validate and modify the AI’s results based on their expertise (65). Research indicates that visual explanations are an effective validation method (66), as 60% of participants felt more confident in decisions made with AI after seeing decision heat maps (67). In complex or atypical circumstances, though, AI models are not always reliable. It is interesting that the regions identified by saliency maps show significant variation and are not always the same as those used in medical practice to define disease severity (68). If the AI output contradicts domain knowledge, the domain knowledge should take precedence and override the model output, based on the clinician’s professional experiences.

The existing approaches to explainable AI have several limitations, including limited cross-center and cross-device validation and small sample sizes (9,63). These gaps must be addressed before widespread clinical use. However, explainable AI is another pivotal step in the future of human-machine collaboration in the field of congenital alveolar cleft management (4,66,69).

The “double-edged sword” of algorithms: bias, generalizability, and ethics

While AI offers promise in the diagnosis and treatment of congenital alveolar clefts, especially in conjunction with 3D image analysis, there are several challenges to overcome in its clinical application, such as algorithmic bias, limited generalizability, and ethical and legal issues.

Algorithmic bias: model bias caused by training data (different populations, different equipment)

The quality and diversity of training data are crucial for the performance of AI algorithms (70). The training sets of patient models for congenital alveolar clefts may be biased depending on the small population or small geographical region, making it difficult to apply to different clinical scenarios. The other point of data heterogeneity in 3D image analysis is the use of various imaging devices and acquisition parameters. A model developed from CBCT images of 194 patients with CLP had a high level of precision, but must be tested in different equipment types and patient populations (4). Mitigation strategies include improving data sets by adding more multicenter collaborations, more sample diversity, and standardizing data acquisition and data pre-processing.

Challenges of General Data Protection Regulation (GDPR) in cross-center studies

A major challenge of the AI-assisted studies of congenital alveolar clefts is the ability to collaborate across centers and share 3D imaging data and clinical information under strict regulatory frameworks such as the Health Insurance Portability and Accountability Act (HIPAA) and the GDPR (71). While these safeguards keep patient data safe, they also restrict collaborations between institutions because of their compliance demands. The GDPR is based on informed consent and data portability, which have the opposite effect to HIPAA and also restrict data sharing and limit the generalizability of models (72). Furthermore, regulatory needs differ across regions, making the global rollout of AI models even more complex. Better data encryption, anonymization, and access are important in ensuring patient privacy, enabling research innovation, and establishing clear consent processes (73). In addition to technical advancements, algorithmic bias, privacy laws, and ethical considerations need to be addressed when using AI in the treatment of congenital alveolar clefts. The development and maintenance of multiple high-quality datasets and data governance are fundamental needs for unlocking the clinical potential of these tools (4).

Beyond the Dice coefficient: from technical validation to clinical benefits

Previous studies investigating the application of AI for the diagnosis and treatment of alveolar clefts have used technical performance metrics such as the Dice coefficient, precision, and recall. These metrics of agreement between model predictions and manual annotations do not fully characterize a model’s role in clinical care. There is a gap between technical accuracy and clinical effectiveness.

Technical validation is usually performed on a standardized dataset in controlled annotation conditions. One study (4) used a deep-learning algorithm on 3D surface models of the cleft lips and palates generated from CBCT images to achieve an F1 score of 0.817, a precision of 0.823, a recall of 0.816, and classification accuracy ranging from 97.4% to 100%, demonstrating strong recognition and classification ability and the automatic extraction of morphological characteristics of the defects for their severity grade.

However, clinical effectiveness requires more than numerical accuracy; model output must be of value to clinical decision-making. The difference was clearly identified by a CLP-Net model applied to automatic localization of CLP-related standard planes from 3D ultrasound images of first-trimester fetuses, with a localization time reduction of 31.3% and 38.9% for senior and junior radiologists, respectively, and a visual acceptance rate of 95% (74).

Beyond the Dice coefficient, the value of model outputs to decision making, surgical planning, and clinical outcomes needs to be evaluated to determine clinical benefits. Technical precision alone is not enough to be effective in the clinical environment: explainability, usability, and integration into workflow are also relevant. The summary of the existing AI models used for alveolar cleft and CLP is shown in Table 2. Future frameworks should investigate clinical outcomes such as the success of bone grafts, the rate of canine eruption, closing of the oral-nasal fistula, long-term periodontal attachment level, and the incidence of secondary alveolar bone grafting to determine the clinical significance of using AI in congenital alveolar cleft repair.

Future prospects: the dawn of alveolar cleft “digital twin”

Deep learning-based automatic segmentation and morphological measurement of CBCT images have established the foundation for clinical use. However, generative AI and digital twin technologies are still in early stages and represent the long-term vision for development.

Generative AI: from data augmentation to surgical simulation

Generative AI, specifically generative adversarial networks (GANs), has shown great promise in the field of congenital deformity research. In addition to creativity, GenAI provides high-fidelity data synthesis (75). Limited datasets can be used to model the latent distributions of anatomical features, and these models can generate realistic CBCT volumes, augmenting training repositories and improving the robustness of downstream segmentation algorithms (76). This goes beyond predictive modeling, where generative AI simulates postoperative changes as a virtual ‘sandbox’ for optimizing surgical protocols before the surgery.

Generation of high-fidelity CBCT data: solving the problem of small sample size and improving model generalizability

High-quality 3D CBCT imaging of patients presenting with congenital alveolar cleft is challenging to acquire at a large scale, and deep learning models trained on limited datasets may exhibit limited generalizability. Generative AI solves this problem by learning from small, real-world datasets and creating high-fidelity 3D images to create a larger dataset of images to learn from (77). In 3D-DGGAN, feature codes of real images are added to the synthetic images, and a multiscale discriminator is designed at three different scales, including the volume level, the consecutive-slice level, and the random-slice level, which enhances the generated image quality and alleviates the limitation imposed by small sample sizes in training (78). Such generative CBCT data also serve as a richer source for downstream model training, thereby enhancing the accuracy of tasks such as alveolar cleft segmentation, reconstruction, and diagnosis, and providing a stronger data basis for clinical diagnosis and treatment planning (79).

Computer-simulated surgery: generative AI predicts the personalized postoperative results of different surgery protocols

There is no standard surgical procedure for a congenital alveolar cleft, and postoperative results depend on the anatomical variation and surgical design. Generative AI can simulate the impact of various protocols on a patient’s 3D morphology and predict postoperative morphology and functional outcomes by training on large datasets of preoperative and postoperative images, which can aid in personalized surgical planning (80). By combining AI-driven simulations with clinical expertise (81), a framework for human-machine co-creation enables physicians to augment the AI’s analytical capabilities without relying only on automated recommendations, ensuring that AI tools remain valuable while human expertise remains essential (82). In this regard, generative AI can be used in two ways for the care of congenital alveolar clefts, each complementing and enhancing the other: high-fidelity CBCT data to overcome limitations in training data, and computer-simulated surgery to support personalized planning of preoperative steps. Further work is needed to optimize model performance and ensure safety and effectiveness in complex clinical environments (83).

Self-supervised learning and federated learning

Two significant recent advances in medical AI are self-supervised learning and federated learning, which address two distinct challenges in the clinical application of AI to congenital alveolar cleft care. Self-supervised learning reduces dependence on large volumes of manually annotated data by training on unannotated images via designed pretraining tasks, allowing models to learn structural and texture features independently (84). This is especially important for research on CLP, which is a relatively uncommon disorder, and for which quality annotation is also time-consuming and prone to inter-rater variability. Self-supervised methods benefit from larger, more abundant pools of unannotated CT or ultrasound images, leading to improved model performance on downstream tasks without requiring a proportional increase in the number of labeled images (85). Federated learning addresses the problem of data being unavailable for sharing since it is sensitive and regulated by privacy policies among institutions (86). In contrast to centralized learning, federated learning distributes data across different sites and trains the model locally, sharing only model parameters or gradients, so as to allow for multicenter collaboration without sharing raw images (87). This approach enhances the appropriateness of the models for various types of equipment and patient groups, for effective early diagnosis and assessment of CLP. Their combined application provides complementary approaches to address the data scarcity that slows the progress of AI and enables further progress to accurately and efficiently characterize the alveolar cleft in 3D for diagnosis and treatment planning.

Ultimate goal: “digital twin” of patients with alveolar cleft

The extensive use of AI and 3D image analysis has made the digital twin concept an achievable goal for the diagnosis and treatment of alveolar clefts. Multiscale, full-cycle simulation and condition management can be achieved by using a digital twin that combines multimodal patient information to generate a high-fidelity dynamic simulation model that incorporates anatomical morphology, biomechanical behavior, and molecular biology (88). It can overcome the previous imaging-based diagnosis, which can only present a single time point of the deformities, such as the alveolar cleft.

A digital twin differs from traditional 3D morphological models, as it incorporates genetic information, morphology, and biomechanics in a multilayered dynamic model (89). It can characterize alveolar bone defect features, simulate interactions between postoperative materials and bone tissue, predict the stability of alveolar bone grafts, and integrate genetic data to model the biological responses related to bone healing through 3D surface reconstruction and morphological classification. This system is able to provide support for the entire process of personalized handling, from prenatal risk prediction through to simulation before surgery, and dynamic monitoring after surgery. Combined genetic and radiological data can identify fetal developmental risk at an early stage, and postoperative biomechanical updates enable real-time monitoring of bone growth to ensure proper treatment and adjust the treatment plan.

Several significant challenges currently limit clinical translation. Multi-modality data integration is one of the fundamental problems, as the heterogeneous data types, incompatible tool interfaces, and data streams create persistent difficulties (90,91). The different scales of spatiotemporal resolution of genomic data and CBCT-based morphology require inter-scale modeling approaches that can unify biological and anatomical analyses (92). Some studies have attempted to address this by projecting genomic variation onto CBCT-displayed bone microstructural characteristics to predict individual bone-healing potential (90), and by constructing universal data models and API interfaces to support multi-source data aggregation (92,93).

Another challenge is biomechanical model integration, because the simulation of stress distribution in bone grafts to masticatory force should include both the mechanical load and the dynamic biological feedback, but most current models only consider mechanical load (94,95). To enhance construction efficiency, it has been proposed that using open-source end-to-end pipelines from CT images to FEA models can be developed (96). Paired collections of mechanistic models and machine learning have been suggested to simulate bone resorption and osteogenesis signaling pathways induced by mechanical stimuli (97). The other limitation is the computational cost and time required. Model development and real-time data processing are complex and require resources and multidisciplinary cooperation (98), and continuous dynamics capture for real-time clinical application is limited by digital hardware (99).

Although digital twinning has proven its capability in the surgical and perioperative environment, it has not yet been possible in resource-limited settings (100). Despite these challenges, continued technical advances are expected to make digital twinning an important tool in alveolar cleft care, moving away from empirical care towards the precision medicine of the future and, finally, towards full-cycle patient management.

Conclusions

AI has demonstrated considerable promise in the field of congenital alveolar cleft management, but there are certain clinical limitations, such as the generalizability of the algorithms, the lack of model explainability, data privacy issues, and the need for frameworks to assess outcomes. Seamless clinical integration is challenging because most studies are retrospective and have small sample sizes and lack external validation (101-103).

Future research must shift from technical validation toward clinical-prognosis-driven priorities, with prospective studies assessing whether AI tools yield improvements in bone graft survival, dental arch continuity, and patient-reported outcomes. The major implementation pathways are standardized databases, federated learning, lightweight models, and AI training.

There is long-term potential for generative AI and digital twin technologies for full-cycle personalized management from the preoperative simulation through the postoperative monitoring. Coordinated progress across technology, data governance, regulation, and clinical practice will ultimately determine whether AI becomes a reliable driver of precision medicine in congenital alveolar cleft care.

Acknowledgments

The language editing and proofreading of this manuscript were kindly provided by Kanlu Paper Editing.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0268/rc

Peer Review File: Available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0268/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tp.amegroups.com/article/view/10.21037/tp-2026-0268/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Wu Q, Li Z, Zhang Y, et al. Dental caries and periodontitis risk factors in cleft lip and palate patients. Front Pediatr 2022;10:1092809. [Crossref] [PubMed]
2. Lu J, Wang Z, Zhang H, et al. Bone Graft Materials for Alveolar Bone Defects in Orthodontic Tooth Movement. Tissue Eng Part B Rev 2022;28:35-51. [Crossref] [PubMed]
3. Fernandes LR, Arce C, Martinho G, et al. Accuracy, Reliability, and Repeatability of a Novel Artificial Intelligence Algorithm Converting Two-Dimensional Radiographs to Three-Dimensional Bone Models for Total Knee Arthroplasty. J Arthroplasty 2023;38:2032-6.
4. Miranda F, Choudhari V, Barone S, et al. Interpretable artificial intelligence for classification of alveolar bone defect in patients with cleft lip and palate. Sci Rep 2023;13:15861. [Crossref] [PubMed]
5. Kurt-Bayrakdar S, Bayrakdar İŞ, Yavuz MB, et al. Detection of periodontal bone loss patterns and furcation defects from panoramic radiographs using deep learning algorithm: a retrospective study. BMC Oral Health 2024;24:155. [Crossref] [PubMed]
6. Chatzopoulos GS, Koidou VP, Tsalikis L, et al. Clinical Applications of Artificial Intelligence in Periodontology: A Scoping Review. Medicina (Kaunas) 2025;61:1066. [Crossref] [PubMed]
7. Xu Y, Quan R, Xu W, et al. Advances in Medical Image Segmentation: A Comprehensive Review of Traditional, Deep Learning and Hybrid Approaches. Bioengineering (Basel) 2024;11:1034. [Crossref] [PubMed]
8. Jin J, Yang S, Tong J, et al. Slim UNETR++: A lightweight 3D medical image segmentation network for medical image analysis. Med Biol Eng Comput 2025;63:3123-37. [Crossref] [PubMed]
9. Fujii Y, Sugiyama-Tamura T, Sugisaki R, et al. New Assessment Method of Alveolar Bone Grafting Using Automatic Registration and AI-based Segmentation. J Craniofac Surg 2024; Epub ahead of print. [Crossref]
10. Wang X, Alqahtani KA, Van den Bogaert T, et al. Convolutional neural network for automated tooth segmentation on intraoral scans. BMC Oral Health 2024;24:804. [Crossref] [PubMed]
11. Hao J, Liao W, Zhang YL, et al. Toward Clinically Applicable 3-Dimensional Tooth Segmentation via Deep Learning. J Dent Res 2022;101:304-11. [Crossref] [PubMed]
12. da Andrade-Bortoletto MFS, Jindanil T, Fontenele RC, et al. Comparison of AI-Powered Tools for CBCT-Based Mandibular Incisive Canal Segmentation: A Validation Study. Clin Oral Implants Res 2025;36:1086-94. [Crossref] [PubMed]
13. HeDUdupaJKTongYPredicting the effort required to manually mend auto-segmentations.medRxiv 2024;2024.06.12.24308779.
14. Hansen S, Gautam S, Salahuddin SA, et al. ADNet++: A few-shot learning framework for multi-class medical image volume segmentation with uncertainty-guided feature refinement. Med Image Anal 2023;89:102870. [Crossref] [PubMed]
15. Hegyi A, Somodi K, Pintér C, et al. Automatic segmentation of dental cone-beam computed tomography scans using a deep learning framework. Orv Hetil 2024;165:1242-51. [Crossref] [PubMed]
16. Qiu B, van der Wel H, Kraeima J, et al. Mandible Segmentation of Dental CBCT Scans Affected by Metal Artifacts Using Coarse-to-Fine Learning Model. J Pers Med 2021;11:560. [Crossref] [PubMed]
17. Yuan N, Rao S, Chen Q, et al. Head and neck synthetic CT generated from ultra-low-dose cone-beam CT following Image Gently Protocol using deep neural network. Med Phys 2022;49:3263-77. [Crossref] [PubMed]
18. Liu Y, Dwivedi G, Boussaid F, et al. Inflating 2D convolution weights for efficient generation of 3D medical images. Comput Methods Programs Biomed 2023;240:107685. [Crossref] [PubMed]
19. Zhu Z, Liu E, Su Z, et al. Three-Dimensional Lumbosacral Reconstruction by An Artificial Intelligence-Based Automated MR Image Segmentation for Selecting the Approach of Percutaneous Endoscopic Lumbar Discectomy. Pain Physician 2024;27:E245-54.
20. Coroama LG, Diosan L, Telecan T, et al. A Light, 3D UNet-based Architecture for Fully Automatic Segmentation of Prostate Lesions from T2-MRI Images. Curr Med Imaging 2023; Epub ahead of print. [Crossref]
21. Coroamă DM, Dioșan L, Telecan T, et al. Fully automated bladder tumor segmentation from T2 MRI images using 3D U-Net algorithm. Front Oncol 2023;13:1096136. [Crossref] [PubMed]
22. Chou PY, Denadai R, Hallac RR, et al. Comparative Volume Analysis of Alveolar Defects by 3D Simulation. J Clin Med 2019;8:1401. [Crossref] [PubMed]
23. Lo HY, Leung PH, Su YX, et al. AI-driven CBCT segmentation and 3D modeling of the anterior surface of maxilla for computer-assisted surgery: a comparison of multiple algorithms. J Craniomaxillofac Surg 2025;53:1683-90. [Crossref] [PubMed]
24. Liu H, Xi X, Liu D. Three-dimensional (3D) quantitative evaluation of the morphological changes of the upper anterior alveolar bone after retraction of a maxillary incisor. BMC Oral Health 2023;23:295. [Crossref] [PubMed]
25. Vicente A, Wiedel AP, Becker M, et al. Quantitative assessment of cleft volume and evaluation of cleft's impact on adjacent anatomical structures using CBCT imaging. Oral Radiol 2024;40:295-303. [Crossref] [PubMed]
26. Padwa BL, Tio P, Garkhail P, et al. Predictors of Outcomes in 900 Alveolar Bone Grafts. Plast Reconstr Surg 2024;154:605-14. [Crossref] [PubMed]
27. Riechelmann H, Widmann G, Kofler B, et al. Nasal Floor Asymmetry Is Associated With Nasal Obstruction. J Oral Maxillofac Surg 2020;78:1833.e1-9.
28. Wojtyńska E, Bączkowski B, Przybyłowska D, et al. A multidisciplinary treatment of patients with craniofacial disorders. Own experience. Dev Period Med 2015;19:464-70.
29. Lee KH, Lee RW, Yun JS, et al. Automated Diagnosis of Knee Osteoarthritis Using ResNet101 on a DEEP:PHI: Leveraging a No-Code AI Platform for Efficient and Accurate Medical Image Analysis. Diagnostics (Basel) 2024;14:2451. [Crossref] [PubMed]
30. Zhang DZ, Xiao WL, Zhou R, et al. Evaluation of Bone Height and Bone Mineral Density Using Cone Beam Computed Tomography After Secondary Bone Graft in Alveolar Cleft. J Craniofac Surg 2015;26:1463-6. [Crossref] [PubMed]
31. Iwai T, Hoshi M, Oebisu N, et al. Prediction of Risk Factors for Pathological Fracture After Bone Tumor Biopsy Using Finite Element Analysis. Cancer Manag Res 2021;13:3849-56. [Crossref] [PubMed]
32. Muhib F, Williams KE, LaBelle SA, et al. Development of a subject-specific finite element analysis workflow to assess local biomechanics during segmental bone defect healing. J Mech Behav Biomed Mater 2025;169:107065. [Crossref] [PubMed]
33. Perier-Metz C, Cipitria A, Hutmacher DW, et al. An in silico model predicts the impact of scaffold design in large bone defect regeneration. Acta Biomater 2022;145:329-41. [Crossref] [PubMed]
34. Borgiani E, Duda GN, Willie BM, et al. Bone morphogenetic protein 2-induced cellular chemotaxis drives tissue patterning during critical-sized bone defect healing: an in silico study. Biomech Model Mechanobiol 2021;20:1627-44. [Crossref] [PubMed]
35. Klunder LS, Merema BBJ, Matthews-Brzozowski AZ, et al. 3D virtual surgical planning in patients with bilateral cleft lip and palate undergoing premaxilla osteotomy combined with secondary alveolar bone grafting: a retrospective accuracy analysis. J Craniomaxillofac Surg 2025;53:1904-10. [Crossref] [PubMed]
36. Sue GR. Artificial Intelligence for Plastic Surgeons. Plast Reconstr Surg Glob Open 2023;11:e5057. [Crossref] [PubMed]
37. Liu Y, Xie R, Wang L, et al. Fully automatic AI segmentation of oral surgery-related tissues based on cone beam computed tomography images. Int J Oral Sci 2024;16:34. [Crossref] [PubMed]
38. Gialain IO, Pinhata-Baptista OH, Cavalcanti MGP, et al. Computer-Aided Design/Computer-Aided Manufacturing Milling of Allogeneic Blocks Following Three-Dimensional Maxillofacial Graft Planning. J Craniofac Surg 2019;30:e413-5. [Crossref] [PubMed]
39. Kanumilli SLD, Kosuru BP, Shaukat F, et al. Advancements and Applications of Three-dimensional Printing Technology in Surgery. J Med Phys 2024;49:319-25. [Crossref] [PubMed]
40. Miyoshi N. Use of AI in Diagnostic Imaging and Future Prospects. JMA J 2025;8:198-203. [Crossref] [PubMed]
41. Almoammar KA. Harnessing the Power of Artificial Intelligence in Cleft Lip and Palate: An In-Depth Analysis from Diagnosis to Treatment, a Comprehensive Review. Children (Basel) 2024;11:140. [Crossref] [PubMed]
42. Forer O, Cohen E, Ben-Bassat Y, et al. Orthodontic Treatment of Patients With Clefts: Satisfaction and Psychological Aspects. Cleft Palate Craniofac J 2023;60:1149-56. [Crossref] [PubMed]
43. Patmon D, Carlson A, Girotto J. Racial Disparities in the Timing of Alveolar Bone Grafting. Cleft Palate Craniofac J 2023;60:1207-10. [Crossref] [PubMed]
44. Huqh MZU, Abdullah JY, Al-Rawas M, et al. Development of Artificial Neural Network-Based Prediction Model for Evaluation of Maxillary Arch Growth in Children with Complete Unilateral Cleft Lip and Palate. Diagnostics (Basel) 2023;13:3025. [Crossref] [PubMed]
45. Gu Y, Yan C, Yan Z, et al. Evaluation of the clinical efficacy and safety of modified alveolar cleft bone graft with cone-beam CT digital imaging in children. Transl Pediatr 2022;11:1140-8. [Crossref] [PubMed]
46. Hou Z, Wang Y, Feng C, et al. Comprehensive overview of biomaterials for bone regeneration in cleft lip and palate. J Mater Chem B 2025;13:12402-22. [Crossref] [PubMed]
47. Wei S, Zhang G, Wu Q, et al. Osteogenesis rate and success of secondary alveolar bone grafting: A 3D volumetric prognostic value. J Plast Reconstr Aesthet Surg 2025;108:168-75. [Crossref] [PubMed]
48. Hong Y, Wang H, Zhang Q, et al. Machine Learning and Deep Learning Hybrid Approach Based on Muscle Imaging Features for Diagnosis of Esophageal Cancer. Diagnostics (Basel) 2025;15:1730. [Crossref] [PubMed]
49. Sousa JV, Matos P, Silva F, et al. Single Modality vs. Multimodality: What Works Best for Lung Cancer Screening? Sensors (Basel) 2023;23:5597.
50. Wójcicki P, Koźlik MJ, Wójcicka K. Genetic Factors in Selected Complex Congenital Malformations with Cleft Defect. Adv Clin Exp Med 2016;25:977-87. [Crossref] [PubMed]
51. Liang S, Chen T, Ma J, et al. Identification of mild cognitive impairment using multimodal 3D imaging data and graph convolutional networks. Phys Med Biol 2024;
52. Park SW, Yeo NY, Kim Y, et al. Deep learning application for the classification of Alzheimer's disease using (18)F-flortaucipir (AV-1451) tau positron emission tomography. Sci Rep 2023;13:8096. [Crossref] [PubMed]
53. Haq IU, Mhamed M, Al-Harbi M, et al. Advancements in Medical Radiology Through Multimodal Machine Learning: A Comprehensive Overview. Bioengineering (Basel) 2025;12:477. [Crossref] [PubMed]
54. Huang X, Han F, Chen YF, et al. Bibliometric analysis of the application of artificial intelligence in orthopedic imaging. Quant Imaging Med Surg 2025;15:3993-4013. [Crossref] [PubMed]
55. Yang G, Guo X, Zhang H, et al. 3D ShiftBTS: Shift Operation for 3D Multimodal Brain Tumor Segmentation. IEEE J Biomed Health Inform 2025;29:6713-26. [Crossref] [PubMed]
56. Chen S, Liu B, Yin N, et al. In-Depth Volumetric Analysis of Alveolar Cleft Defects Using Three-Dimensionally Printed Models. J Craniofac Surg 2020;31:1790-2. [Crossref] [PubMed]
57. Syed S, Ehsan L, Shrivastava A, et al. Artificial Intelligence-based Analytics for Diagnosis of Small Bowel Enteropathies and Black Box Feature Detection. J Pediatr Gastroenterol Nutr 2021;72:833-41. [Crossref] [PubMed]
58. Yan X, Duan F, Chen L, et al. A Multimodal MRI-Based Model for Colorectal Liver Metastasis Prediction: Integrating Radiomics, Deep Learning, and Clinical Features with SHAP Interpretation. Curr Oncol 2025;32:431. [Crossref] [PubMed]
59. Rosero K, Salman AN, Harrison LM, et al. Deep Learning-Based Assessment of Lip Symmetry for Patients With Repaired Cleft Lip. Cleft Palate Craniofac J 2025;62:289-99. [Crossref] [PubMed]
60. Tekkesinoglu S, Pudas S. Explaining graph convolutional network predictions for clinicians-An explainable AI approach to Alzheimer's disease classification. Front Artif Intell 2023;6:1334613. [Crossref] [PubMed]
61. Aradya A, Sravani K, Ravi MB, et al. Artificial intelligence for maxillofacial prosthodontics: A technological shift in craniofacial rehabilitation- a scoping review. J Oral Biol Craniofac Res 2025;15:1749-66. [Crossref] [PubMed]
62. Jiang W, Pang H, Deng S, et al. Artificial intelligence-based prenatal ultrasound for diagnosing fetal lip and palate during the second trimester. Quant Imaging Med Surg 2025;15:7497-509. [Crossref] [PubMed]
63. Miller DD. Can AI help with the hardest thing: pro health behavior change. NPJ Cardiovasc Health 2026;3:3. [Crossref] [PubMed]
64. The new EU General Data Protection Regulation: what the radiologist should know. Insights Imaging 2017;8:295-9.
65. Staunton C, Slokenberga S, Parziale A, et al. Appropriate Safeguards and Article 89 of the GDPR: Considerations for Biobank, Databank and Genetic Research. Front Genet 2022;13:719317. [Crossref] [PubMed]
66. Wang C, Zhang J, Lassi N, et al. Privacy Protection in Using Artificial Intelligence for Healthcare: Chinese Regulation in Comparative Perspective. Healthcare (Basel) 2022;10:1878. [Crossref] [PubMed]
67. He G, Li Z, Zhu Z, et al. CLP-Net: an advanced artificial intelligence technique for localizing standard planes of cleft lip and palate by three-dimensional ultrasound in the first trimester. BMC Pregnancy Childbirth 2025;25:10. [Crossref] [PubMed]
68. Arora V, Thabane A, Parpia S, et al. Generative artificial intelligence models outperform students on divergent and convergent thinking assessments. Sci Rep 2025;15:36987. [Crossref] [PubMed]
69. Koetzier LR, Wu J, Mastrodicasa D, et al. Generating Synthetic Data for Medical Imaging. Radiology 2024;312:e232471. [Crossref] [PubMed]
70. Porkodi SP, Sarada V, Maik V, et al. Generic image application using GANs (Generative Adversarial Networks): A Review. Evol Syst (Berl) 2022. [Epub ahead of print]. doi: 10.1007/s12530-022-09464-y.
71. Kim J, Li Y, Shin BS. 3D-DGGAN: A Data-Guided Generative Adversarial Network for High Fidelity in Medical Image Generation. IEEE J Biomed Health Inform 2024;28:2904-15. [Crossref] [PubMed]
72. Zhang Y, Yue N, Su MY, et al. Improving CBCT quality to CT level using deep learning with generative adversarial network. Med Phys 2021;48:2816-26. [Crossref] [PubMed]
73. Miragall MF, Knoedler S, Kauke-Navarro M, et al. Face the Future-Artificial Intelligence in Oral and Maxillofacial Surgery. J Clin Med 2023;12:6843. [Crossref] [PubMed]
74. Chompunuch S, Lubart T. AI as a Helper: Leveraging Generative AI Tools Across Common Parts of the Creative Process. J Intell 2025;13:57. [Crossref] [PubMed]
75. McGuire J, De Cremer D, Van de Cruys T. Establishing the importance of co-creation and self-efficacy in creative collaboration with artificial intelligence. Sci Rep 2024;14:18525. [Crossref] [PubMed]
76. Rafner J, Beaty RE, Kaufman JC, et al. Creativity in the age of generative AI. Nat Hum Behav 2023;7:1836-8. [Crossref] [PubMed]
77. Xu K, You K, Zhu B, et al. Masked Modeling-Based Ultrasound Image Classification via Self-Supervised Learning. IEEE Open J Eng Med Biol 2024;5:226-37. [Crossref] [PubMed]
78. Peng J, Wang P, Pedersoli M, et al. Boundary-aware information maximization for self-supervised medical image segmentation. Med Image Anal 2024;94:103150. [Crossref] [PubMed]
79. Loftus TJ, Ruppert MM, Shickel B, et al. Federated learning for preserving data privacy in collaborative healthcare research. Digit Health 2022;8:20552076221134455. [Crossref] [PubMed]
80. Wu S, Chen J, Nie X, et al. Global prototype distillation for heterogeneous federated learning. Sci Rep 2024;14:12057. [Crossref] [PubMed]
81. Venkatesh KP, Raza MM, Kvedar JC. Health digital twins as tools for precision medicine: Considerations for computation, implementation, and regulation. NPJ Digit Med 2022;5:150. [Crossref] [PubMed]
82. Mardinoglu A, Turkez H, Shong M, et al. Longitudinal big biological data in the AI era. Mol Syst Biol 2025;21:1147-65. [Crossref] [PubMed]
83. Shen S, Qi W, Liu X, et al. From virtual to reality: innovative practices of digital twins in tumor therapy. J Transl Med 2025;23:348. [Crossref] [PubMed]
84. Mosquera-Lopez C, Jacobs PG. Digital twins and artificial intelligence in metabolic disease research. Trends Endocrinol Metab 2024;35:549-57. [Crossref] [PubMed]
85. Vallée A. Digital Twins for Personalized Medicine Require Epidemiological Data and Mathematical Modeling: Viewpoint. J Med Internet Res 2025;27:e72411. [Crossref] [PubMed]
86. Sadée C, Testa S, Barba T, et al. Medical digital twins: enabling precision medicine and medical artificial intelligence. Lancet Digit Health 2025;7:100864. [Crossref] [PubMed]
87. Nikolova N, Raykovska M, Petkov N, et al. The Integration of Micro-CT Imaging and Finite Element Simulations for Modelling Tooth-Inlay Systems for Mechanical Stress Analysis: A Preliminary Study. J Funct Biomater 2025;16:267. [Crossref] [PubMed]
88. Kadry K, Gupta S, Nezami FR, et al. Probing the limits and capabilities of diffusion models for the anatomic editing of digital twins. NPJ Digit Med 2024;7:354. [Crossref] [PubMed]
89. Ariyanfar A, Bahrami M, Klein K, et al. Fast automated creation of digital twins for virtual mechanical testing of ovine fractured tibiae. Comput Biol Med 2025;192:110268. [Crossref] [PubMed]
90. De Domenico M, Allegri L, Caldarelli G, et al. Challenges and opportunities for digital twins in precision medicine from a complex systems perspective. NPJ Digit Med 2025;8:37. [Crossref] [PubMed]
91. Checcucci E, Oing C, Amparore D, et al. Digital Twins in Urological Oncology: Precise Treatment Planning via Complex Modeling. Eur Urol Oncol 2025;8:1-4. [Crossref] [PubMed]
92. Chen H, Yang J, Chen J, et al. Continuous-time digital twin with analog memristive neural ordinary differential equation solver. Sci Adv 2025;11:eadr7571. [Crossref] [PubMed]
93. Mekki YM, Luijten G, Hagert E, et al. Digital twins for the era of personalized surgery. NPJ Digit Med 2025;8:283. [Crossref] [PubMed]
94. Bhat M, Rabindranath M, Chara BS, et al. Artificial intelligence, machine learning, and deep learning in liver transplantation. J Hepatol 2023;78:1216-33. [Crossref] [PubMed]
95. Subhan S, Malik J, Haq AU, et al. Role of Artificial Intelligence and Machine Learning in Interventional Cardiology. Curr Probl Cardiol 2023;48:101698. [Crossref] [PubMed]
96. Alsallal M, Habeeb MS, Vaghela K, et al. Artificial intelligence in gastric cancer: a systematic review of machine learning and deep learning applications. Abdom Radiol (NY) 2026;51:1694-710. [Crossref] [PubMed]
97. Vicente A, Hung KF, Xu Z, et al. Deep learning for automated alveolar cleft segmentation and bone graft volume estimation in cone-beam computed tomography imaging: a multicenter study. Oral Surg Oral Med Oral Pathol Oral Radiol 2026;141:400-8. [Crossref] [PubMed]
98. Shankar VA, Snyder-Warwick A, Skolnick GB, et al. Incidence of Palatal Fistula at Time of Secondary Alveolar Cleft Reconstruction. Cleft Palate Craniofac J 2018;55:999-1005. [Crossref] [PubMed]
99. Ruslin M, Stevanie C, Tajrin A. Advanced Strategies in the Repair of Alveolar Cleft Defects: Biological Approaches, Biomaterials, and Future Perspectives. Prog Mol Subcell Biol 2026;63:209-35. [Crossref] [PubMed]
100. El Deeb M, Messer LB, Lehnert MW, et al. Canine eruption into grafted bone in maxillary alveolar cleft defects. Cleft Palate J 1982;19:9-16.
101. Hogan L, Shand JM, Heggie AA, et al. Canine eruption into grafted alveolar clefts: a retrospective study. Aust Dent J 2003;48:119-24. [Crossref] [PubMed]
102. Din M, Daga K, Saoud J, et al. Clinicians' perspectives on the use of artificial intelligence to triage MRI brain scans. Eur J Radiol 2025;183:111921. [Crossref] [PubMed]
103. Fumero F, Sigut J, Estévez J, et al. Systematic application of saliency maps to explain the decisions of convolutional neural networks for glaucoma diagnosis based on disc and cup geometry. Biomed Phys Eng Express 2025;