AEAsghar Ebadifar

Asghar Ebadifar

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Asghar Ebadifar has a total of 22 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Artificial Intelligence-based multi-stage pipeline for detecting caries and restorations in pediatric primary molars from intraoral photographs

    by Shahab Kavousinejad, Parisa Motie, Yasaman Daghighi, Alireza Norouzi, Hossein Rabbani, Nazila Biglar, Asghar Ebadifar, Mohammad Behnaz

  • Top Cited Papers

    Artificial Intelligence–Based Multi-Stage System for Automated Angle’s Classification of Malocclusion from Intraoral Images in Orthodontics

    by Shahab Kavousinejad, Sara Ghazanfari, Sara Alsadat Hosseinikhah Manshadi, Nazila Biglar, Golnaz Nahvi, Kazem Dalaie, Mohammad Behnaz, Hoori Mirmohammadsadeghi, Soodeh Tahmasbi, Asghar Ebadifar

    This study presents a multi-stage deep learning pipeline for automated Angle's classification of occlusion using intraoral images in orthodontics. The pipeline integrates three key stages: (1) a binary Occlusion Side Classifier (OSC) to determine whether the input image represents the right or left side of the patient's dentition, (2) side-specific bounding box detection models (MolarBBoxNet-R or MolarBBoxNet-L) to localize the molar region, and (3) a unified classifier (AngleClassifier-R50) to predict Class I, II, or III occlusal relationships. A dataset of 8909 lateral intraoral occlusion images from three orthodontic centers was used, including patients older than 6 years, ensuring the presence of fully erupted first molars. Images were annotated by two experts, with discrepancies resolved through consensus. The pipeline achieved perfect occlusion side classification (accuracy 1.00) and high molar classification accuracy (97.41% internal, 94.3% external), with sensitivity and specificity on the external validation set (383 previously unseen images) of 99.0%/98.9% for Class I, 88.2%/98.7% for Class II, and 97.2%/93.4% for Class III. Grad-CAM visualizations confirmed the model's focus on clinically relevant molar regions, enhancing interpretability. The system processed images in ~ 0.11 s, significantly faster than expert annotation (p < 0.001). By operating solely on intraoral photographs, without radiographs, the approach demonstrates practical potential for general practice and tele-orthodontics. It may reduce diagnostic variability, support large-scale screening, and alleviate clinical triage burden. Future work should expand the framework to incorporate additional occlusal parameters and multimodal inputs such as radiographs or 3D scans for comprehensive orthodontic diagnostics.


  • Deep Learning-Based Prediction of Orthognathic Surgery Need in Skeletal Class II Patients Using Facial Profile Images: An Artificial Intelligence Study

    by Soodeh Tahmasbi, Hadi Shokoofi, Kazem Dalaie, Mohammad Behnaz, Farzad Aghdashi, Farshid Bastami, Asghar Ebadifar, Hoori Mirmohammadsadeghi, Nazila Biglar, Shahab Kavousinejad

    Orthognathic surgery is often required to address moderate to severe skeletal class II malocclusion, a condition that affects both facial aesthetics and function. Traditional diagnosis relies on cephalometric radiographs and expert evaluation, which can be time-consuming and subject to inter-observer variability. The need for AI-assisted initial screening based exclusively on lateral soft tissue profiles is increasingly recognized, particularly for identifying mandibular retrusion relative to the maxilla. This study aimed to develop and evaluate a deep learning-based model for classifying lateral facial profile photographs into surgical (S) and non-surgical (NS) categories. A dataset of 524 standardized profile images from skeletal class II patients was collected from three dental centers. Each image was independently reviewed by a panel of five experts (three orthodontists and two maxillofacial surgeons), with labels assigned based on majority consensus. To minimize visual bias and enhance model performance, images underwent preprocessing involving facial cropping, background removal via U2-Net segmentation, silhouette contour extraction, and resizing to 128 × 128 pixels. A custom convolutional neural network (ContourNet) was developed and optimized using fivefold cross-validation. The model achieved an overall accuracy of 90%, with a precision of 92% for NS and 88% for S, and a recall of 88% for NS and 92% for S. Saliency maps revealed that the model focused on clinically relevant regions, particularly the lower jaw and chin in surgical cases. This deep learning framework demonstrates strong potential as a non-invasive, efficient tool for preliminary screening in orthognathic surgery decision-making. Further external validation on larger and more diverse populations is recommended prior to clinical implementation.


  • Artificial Intelligence Framework for Automated Facial Asymmetry Detection Using Key-Point Analysis and Neural Networks

    by Shahab Kavousinejad, Yasamin Vazirizadeh, Mohammad Behnaz, Asghar Ebadifar, Hoori Mirmohammadsadeghi

    Accurate facial asymmetry assessment is essential in orthodontics, maxillofacial surgery, and plastic surgery. While minor asymmetry is common, severe cases often result from congenital conditions or trauma. Traditional methods struggle to comprehensively quantify asymmetry’s extent and direction. This study developed and compared artificial neural networks (ANN) and Siamese neural networks (SNN) to detect facial asymmetry and determine deviation direction (horizontal/vertical). A dataset of 1200 frontal photographs, annotated by three orthodontists, was used. The MediaPipe model facilitated facial landmark detection and midline alignment. Two approaches were employed: (1) extracting features from facial landmarks and using them to train an ANN, and (2) SNN-based comparison of mirrored facial halves. Exploratory data analysis (EDA) was used to quantify facial asymmetry in both vertical and horizontal dimensions. The ANN and SNN performance was evaluated using accuracy, recall, and F1-score. The SNN outperformed ANN, achieving 97% accuracy and strong agreement with expert evaluations (Cohen’s Kappa: 0.84 for asymmetry detection, 0.73 for horizontal deviation, and 0.80 for vertical asymmetry). The symmetry group showed 96.14% mean similarity, while the asymmetry group had 83.97%. The SNN’s ROC curve yielded an AUC of 0.98, indicating high diagnostic performance. This study demonstrates the potential of AI-driven methods, particularly SNN, for reliable and objective facial asymmetry assessment in clinical settings. Future research should focus on expanding datasets and refining midline alignment to improve accuracy, especially in cases with vertical eye asymmetry.