EfficientNet-B0 Deep Learning Model for Accurate Classification of Intrabony Lesions in Dental Panoramic Radiographs

EfficientNet-B0 Deep Learning Model for Accurate Classification of Intrabony Lesions in Dental Panoramic Radiographs | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article EfficientNet-B0 Deep Learning Model for Accurate Classification of Intrabony Lesions in Dental Panoramic Radiographs Ammar Mohi, Lubna Kareem, Atef A Hassan This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6207574/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Objectives To develop and evaluate an EfficientNet-B0-based deep-learning model for classifying eight types of intrabony lesions on dental panoramic radiographs. Methods A dataset of 833 dental panoramic radiographs from 245 patients was collected between October 2021 and April 2023 at two dental centers. Images were classified into eight categories: Periapical Widening, Condensing Osteitis, Periapical Granuloma, Nil Control, Diffuse Lesion, Periapical Abscess, Pericoronitis, and Radicular Cyst. Data preprocessing included class-weight computation and augmentation of minority classes. The EfficientNet-B0 model was trained for 50 epochs using the Adam optimizer with learning-rate scheduling and mixed-precision training. Results The model achieved 93.04% validation accuracy, 0.9345 precision, 0.9304 recall, and 0.9295 F1-score. Performance analysis demonstrated robust classification across all lesion types, with the highest accuracy in Nil Control and Radicular Cyst identification. Conclusions The EfficientNet-B0 model demonstrates high accuracy in classifying dental intrabony lesions from panoramic radiographs, offering potential for enhanced diagnostic precision in clinical settings. Further validation across diverse clinical environments is recommended to establish a broader applicability. Deep learning Dental radiography Intrabony lesions Panoramic radiographs Computer-aided diagnosis Dental pathology Machine learning Neural networks Image classification Diagnostic imaging Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Accurate diagnosis of intrabony lesions is pivotal in dental practice, as these conditions can significantly impact oral health and overall patient wellbeing 1 , 2 . Intrabony lesions, including periapical abscesses, condensing osteitis, and radicular cysts, often result from chronic infections, trauma, or developmental anomalies. Timely and precise identification of these lesions through radiographic imaging is essential for determining appropriate treatment strategies, preventing disease progression, and enhance patient outcomes 3 . Orthopantomograms (OPGs), or panoramic radiographs, are widely used in dental diagnostics because of their ability to provide a comprehensive view of the maxillofacial region in a single image 4 . However, the interpretation of OPGs requires substantial expertise, and the manual analysis process is subject to inter-observer variability and potential diagnostic errors 5 . These challenges underscore the need for automated, reliable, and efficient diagnostic tools to assist dental professionals in accurately classifying intrabony lesions. In recent years, deep learning (DL) and convolutional neural networks (CNNs) have emerged as transformative technologies in medical imaging, offering unprecedented accuracy and efficiency in image classification and pattern recognition tasks 6 , 7 . Among the various CNN architectures, EfficientNet-B0 has gained prominence for its compound scaling method, which uniformly scales the network depth, width, and resolution, resulting in superior performance with fewer parameters compared to other models 8 , 9 . This efficiency makes EfficientNet-B0 particularly suitable for applications that require high-resolution image analysis such as dental radiology 10 . Despite advancements in DL applications in medical imaging, there remains a gap in the automated classification of diverse intrabony lesions on dental radiographs. Most existing studies have focused on specific conditions or have utilized architectures primarily designed for object detection rather than comprehensive image classification 11 , 12 . To address this gap, the present study aimed to develop and evaluate an EfficientNet-B0-based model for the accurate classification of intrabony lesions in dental X-ray images. By leveraging a robust dataset and advanced preprocessing techniques, this study sought to enhance diagnostic precision and support clinical decision-making in dental radiology. Methods and materials/patients Study Design This retrospective study was meticulously designed in accordance with established dental protocols and ethical standards across multiple private dental clinics and specialized centers. Ethical approval was obtained from the [Al Farahidi University / Collage of Dentistry] to ensure compliance with all regulatory guidelines. Oversight was provided by Two Oral and Maxillofacial (OMF) radiologists to maintain adherence to diagnostic criteria and ethical protocols. The primary objective was to develop and evaluate an EfficientNet-B0-based convolutional neural network (CNN) model for the accurate classification of intrabony lesions in dental X-ray images by leveraging advanced deep learning techniques to enhance diagnostic precision and support clinical decision-making. Eligibility Criteria Orthopantomogram (OPG) panoramic radiographs were collected between October 2021 and April 2023 at the Basamat Private Dental Clinic and Al-Sha'ab Specialized Center. The inclusion criteria were age between 18 and 70 years, absence of previous jaw trauma or fractures, presence of partial or complete edentulous alveolar bone, broad dental inclusion dimensions within the included dew (FOV) of the OPG image, high-resolution images devoid of artifacts, and achieved centric occlusion during OPG exposure with proper head positioning. Exclusion criteria eliminated Patients with a history of facial trauma, dentofacial malformations or anomaly syndromes, developmental bone diseases, osteometabolic disorders, or hormonal diseases affecting bone morphology were excluded. Study Sample A total of 245 patients' OPG images were meticulously screened and included in the analysis, resulting in 833 training and validation samples categorized into eight distinct classes: Periapical Widening, Condensing Osteitis, Periapical Granuloma, Nil Control (no pathological findings), Diffuse Lesion, Periapical Abscess, Pericoronitis, and Radicular Cyst. Each class represents a specific intrabony lesion selected based on its clinical significance and frequency of occurrence in dental radiology. Data Collection Radiographic imaging was performed using a Myray 2D Pro X9 Panoramic Device (FOV 15 × 15 cm; 50–100 kVp; 1–22 mA; 20 µs) from Myray (Cephla, Bologna, Italy). This device was chosen for its high-resolution imaging capabilities, which are essential for detailed evaluation of intra-bony bone morphology, density, and spatial positioning based on anatomical and radiological principles. Each intra-alveolar bone pathology was categorized as radiopaque or radiolucent and further classified according to site, position, and size. Data Preprocessing and Augmentation Data preprocessing and augmentation were performed using a custom PyTorch Dataset class, the DentalXRayDataset, designed to validate directory structures, retrieve class names, and compile lists of image file paths with corresponding labels. Data augmentation techniques were employed using the Albumentation library 13 to enhance the model's generalization capabilities and simulate real-world variability inherent in clinical radiographs. The training data were resized to 224 × 224 pixels, with random rotations within ± 20 °, horizontal flipping with a training ability of 0.5, brightness and contrast adjustments with a probability of 0.2, and normalization on ImageNet statistics. Advanced augmentations, including random 90-degree rotations, Gaussian noise, Gaussian blur, motion blur, median blur, and general blurring, were incorporated to further diversify the training data. To maintain the integrity of the evaluation metrics, the validation data underwent minimal preprocessing, limited to resizing and normalization. Handling Class Imbalance Addressing the inherent class imbalance, particularly with classes such as pericoronitis (24 images), involves computing class weights using the compute_class_weight function from scikit-learn. These weights were integrated into the CrossEntropyLoss function 14 , allowing the model to assign appropriate emphasis to minority classes during training. Additionally, aggressive data augmentation techniques have been applied to augment minority classes, thereby increasing their effective sample size and diversity. Model Architecture EfficientNet-B0 Configuration The classification model was built using the EfficientNet-B0 architecture, selected for its compound scaling method, which uniformly scales the network depth, width, and resolution, achieving superior performance with fewer parameters. The architecture was modified to accommodate the specific requirements of the study by configuring the input layer to accept images with a size of 224 × 224 pixels, replacing the final fully connected layer to match the number of target classes (eight), and incorporating a dropout layer to prevent overfitting. Optimizer and Scheduler Training was conducted over 50 epochs using the Adam optimizer with an initial learning rate of 0.001 and weight decay of 0.0005 to prevent overfitting by penalizing large weights. A StepLR scheduler was utilized, reducing the learning rate by a factor of 0.1 every three epochs to facilitate finer weight adjustments as training progressed, promoting stabilization, and preventing overfitting. Mixed Precision Training and Gradient Scaling To optimize the training efficiency and leverage the computational capabilities of modern GPUs, mixed-precision training was implemented using Torch. cuda. amp. This approach, coupled with gradient scaling via GradScaler, accelerates the training process and reduces memory consumption without compromising the model performance. Training Procedure The training loop involves a systematic process in which each epoch includes a forward pass through the EfficientNet-B0 model to obtain predictions, computation of cross-entropy loss incorporating class weights, backpropagation using mixed-precision training with autocast, and optimizer updates based on scaled gradients. The learning rate scheduler adjusts the learning rate based on a predefined schedule. After each epoch, performance metrics, such as accuracy, precision, recall, F1-score, and confusion matrix, were calculated for the validation set. The state of the model was saved whenever an improvement in the validation accuracy was observed, ensuring the retention of the best-performing model. Evaluation Metrics The model performance was rigorously evaluated using a suite of metrics to ensure a holistic assessment of classification capabilities. The metrics included the accuracy, precision, recall, F1-score, confusion matrix, and specificity. Precision measures the ratio of true positives to the sum of true and false positives, whereas recall (sensitivity) assesses the ratio of true positives to the sum of true positives and false negatives. The F1-score provided a balance between precision and recall. The confusion matrix offered a detailed breakdown of true versus predicted classifications across all classes, and the specificity evaluated the model's ability to correctly identify non-pathological cases. Results The dental X-ray classification model built upon the EfficientNet-B0 architecture demonstrated exceptional performance in classifying intrabony lesions across eight distinct categories. The model was trained and validated using a dataset of 833 images with Fig. 1 showing the class distributions of different classes in the database. Overall Evaluation Metrics As shown in Table 1 , the model achieved a final validation accuracy of 93.04% and a validation loss of 0.2135. In addition, the model demonstrated robust performance across various metrics, including the precision, recall, and F1-score, with values of 0.9345, 0.9304, and 0.9295, respectively. Table 1 Final Evaluation Metrics Metric Value Final Validation Accuracy 93.04% Final Validation Loss 0.2135 Final Precision 0.9345 Final Recall 0.9304 Final F1 Score 0.9295 Class-wise Performance The model's precision, recall, and F1-score metrics underscore its robust performance across all categories. Table 2 details the class-wise metrics, highlighting the high precision and recall values, which indicate the model's proficiency in correctly identifying true positives, while maintaining low rates of false positives and negatives. Table 2 Class-wise Precision, Recall, and F1-Score Class Precision Recall F1-Score Periapical Widening 0.87 0.85 0.86 Condensing Osteitis 0.89 0.92 0.90 Periapical Granuloma 0.90 0.88 0.89 Nil Control 0.95 0.94 0.94 Diffuse Lesion 0.88 0.85 0.86 Periapical Abscess 0.91 0.93 0.92 Pericoronitis 0.93 0.91 0.92 Radicular Cyst 0.94 0.95 0.94 Training Dynamics Table 3 presents a snapshot of the model performance across the selected epochs, illustrating a progressive improvement over 50 epochs. Significant enhancements were evident from epochs 1 to 20, when the model rapidly increased in accuracy and decreased in loss. By epoch 50, the model achieved high validation accuracy with minimal loss, signifying effective learning and generalization. Table 3 Epoch-wise Training and Validation Metrics Epoch Train Loss Train Acc (%) Val Loss Val Acc (%) 1 2.0224 23.89 2.2062 19.81 10 0.9202 69.27 0.9202 69.27 20 0.2130 93.04 0.2130 93.04 30 0.2528 91.24 0.2528 91.24 40 0.2130 93.04 0.2130 93.04 50 0.1998 91.48 0.2135 93.04 The accuracy plots demonstrated a consistent upward trajectory for both training and validation accuracies. The early epochs exhibited rapid improvements, which gradually stabilized as the model approached convergence. The final training and validation accuracies reached 91.48% and 93.04%, respectively, indicating effective learning and generalization. The loss curves exhibited a steady decline in both training and validation losses. The training loss decreases from an initial value of 2.0224 to a final value of 0.1998, whereas the validation loss decreases from 2.2062 to 0.2135. This consistent reduction indicates an improved capacity of the model to minimize prediction errors. The precision, recall, and F1-score plots indicate steady improvement throughout the training process. By epoch 50, the model achieved high-precision recall values, resulting in an F1-score of 0.9295. This convergence suggests that the model effectively balances precision and recall, minimizing both false positives and negatives. The confusion matrix provides a detailed breakdown of the classification performance of the model across all classes. High values along the diagonal and low off-diagonal values demonstrated the model's ability to accurately distinguish between different dental conditions with minimal misclassifications. The learning rate scheduler played a pivotal role in the training process of the model. The learning rate was reduced at predefined intervals (every three epochs). The scheduler facilitates finer weight updates, contributing to the stabilization of training and the prevention of overfitting. This strategic adjustment enhances the model’s ability to converge to an optimal solution. Discussion This study successfully developed and evaluated an EfficientNet-B0-based model for the classification of intrabony lesions in dental X-ray images. Achieving a validation accuracy of 93.04% and a robust F1 score of 0.9295, the model demonstrates strong performance across eight distinct categories, including Periapical Widening, Condensing Osteitis, Periapical Granuloma, Nil Control, Diffuse Lesion, Periapical Abscess, Pericoronitis, and Radicular Cyst. These results underscore the potential of EfficientNet-B0 to enhance the diagnostic accuracy and efficiency in dental radiology. Interpretation of Results The high validation accuracy and F1 score indicated that the EfficientNet-B0 model effectively distinguished between various intra-bony lesions with minimal misclassifications. Notably, the model exhibited superior performance in classes such as Nil Control (precision, 0.95; recall, 0.94) and radicular cysts (precision, 0.94; recall, 0.95), suggesting its proficiency in identifying conditions with distinct radiographic features. Conversely, classes such as periapical widening and diffuse lesions recorded slightly lower precision and recall values (0.87 and 0.88, respectively), which may be attributed to overlapping radiographic characteristics with other lesion types or inherent variability within these conditions. Comparative Analysis with Recent Studies A recent study by Yilmaz et al. (2023) 15 employed YOLOv4 and Faster R-CNN, which are primarily designed for object detection tasks and can localize and classify objects within an image. This study focused on tooth classification, identifying 36 classes, including 32 teeth and four impacted teeth, in a dataset of 1,200 panoramic radiographs. The YOLOv4 model achieved outstanding precision (99.90%), recall (99.18%), and F1 score (99.54%), thereby significantly outperforming the Faster R-CNN method. In contrast, our study utilized EfficientNet-B0, an image classification model optimized for categorizing entire images rather than detecting and localizing individual objects. In addition, Yilmaz et al 15 . utilized a larger dataset of 1,200 images with 36 classes, focusing on individual tooth identification and classification. However, our study used 833 training and validation images across eight intra-bony lesion classes, representing a more complex classification task owing to the nuanced differences between the lesion types. The task of Yilmaz et al 15 . involves the identification and classification of individual teeth, which generally present more distinct and localized features. In contrast, our task involves classifying various types of intra-bony lesions that can exhibit overlapping and subtle radiographic characteristics, increasing the complexity of the classification process. While Yilmaz et al 15 . achieved near-perfect precision and recall in tooth classification tasks, our EfficientNet-B0 model demonstrated strong performance in a more complex classification scenario involving varied lesion types, achieving a precision of 0.9345, recall of 0.9304, and an F1-score of 0.9295. The slight reduction in performance metrics is attributable to the inherent complexity of distinguishing intra-bony lesions, which often requires discerning subtle differences in bone morphology and density, which are less prominent in tooth identification. Kurt et al 16 . utilized YOLOv5, which is designed for real-time object detection and classification, enabling the simultaneous localization and classification of objects within an image. Our study employed EfficientNet-B0, which was optimized for image-classification tasks without explicit localization capabilities. Additionally, Kurt et al. analyzed 1,500 images with 46 classes corresponding to different tooth development stages and compared them to our 833 training and validation images across eight lesion categories. Kurt et al. focused on detecting and classifying tooth development stages, a task that benefits from the precise localization and high sensitivity offered by YOLOv5. In contrast, our study involved classifying intrabony lesions, requiring a model to differentiate between multiple lesion types with overlapping features. Kurt et al 16 . reported a sensitivity of 0.99, precision of 0.72, and an F1 score of 0.84. A high sensitivity indicates excellent true-positive detection, which is crucial for ensuring that the most relevant developmental stages are identified. However, the lower precision suggests a higher rate of false positives compared to our model. Our EfficientNet-B0 model achieved a balanced precision (0.9345) and recall (0.9304), culminating in a robust F1 score (0.9295). This balance minimizes both false positives and false negatives, and enhances the reliability of diagnoses in clinical settings. While YOLOv5 excels in tasks requiring high sensitivity, such as developmental stage detection, EfficientNet-B0 offers a balanced approach that is ideal for complex lesion classification, where both precision and recall are paramount. Our study was part of a larger project that aimed to collect additional data and images covering a wider range of classes and pathologies. The project will implement more advanced techniques and tools to assist specialists in saving time and effort by developing a web-based software application built on this model. Model-Specific Insights EfficientNet-B0 Architecture EfficientNet-B0 is renowned for its compound scaling method, which uniformly scales all dimensions of depth, width, and resolution using a simple yet highly effective compound coefficient. This balanced scaling enables EfficientNet-B0 to achieve superior performance with fewer parameters than the other models, making it both computationally efficient and highly accurate. In dental radiology, where high-resolution images are paramount for an accurate diagnosis, EfficientNet-B0's ability to extract intricate features from detailed radiographs is particularly advantageous. Feature Extraction and Learning Efficiency The architecture of the EfficientNet-B0 model facilitates effective feature extraction through its deep and narrow layers, allowing it to capture both global and local patterns within radiographic images. This capability is crucial for differentiating various intrabony lesions that may exhibit subtle differences in bone density, morphology, and spatial positioning. Additionally, the model's efficient use of computational resources ensures faster training and inference times, making it suitable for real-time clinical applications in which timely decision making is essential. Data Augmentation and Preprocessing The implementation of data augmentation techniques such as rotation, horizontal flipping, and brightness adjustments plays a pivotal role in enhancing the generalization capabilities of the model. These augmentations simulate the variability inherent in clinical radiographs, such as differences in patient positioning and imaging conditions, thereby reducing the risk of overfitting and improving the model robustness across diverse clinical scenarios. Clinical Importance in Management of Different Conditions The ability to accurately classify intrabony lesions has profound implications for the clinical management of dentistry. Each lesion type identified by the model corresponded to a distinct pathological condition that required specific treatment approaches. Periapical Widening and Periapical Abscess are conditions that often result from chronic pulpitis and can lead to significant bone loss and tooth mobility if left untreated. Early and accurate detection enables timely root canal therapy or extraction, preventing further complications, such as bone destruction and systemic infections. Condensing osteitis and radicular cysts present different clinical challenges. Condensing osteitis is typically a response to low-grade inflammation, whereas radicular cysts are developmental lesions that arise from epithelial remnants. Differentiating between these conditions is crucial, as condensing osteitis may resolve with conservative treatment, whereas radicular cysts might require surgical intervention to prevent bone erosion and maintain jaw integrity. Pericoronitis involves inflammation around partially erupted molars, often leading to pain and infection, whereas Diffuse Lesions can indicate a variety of pathologies, including benign and malignant conditions 17 . Accurate classification assists in determining the appropriate intervention, whether it is antimicrobial therapy, removal of the offending tooth in pericoronitis, or further diagnostic imaging and biopsy for diffuse lesions. Nil Control classifications are equally important, as correctly identifying images with no pathological findings is essential to avoid unnecessary treatment and alleviate patient anxiety. Ensuring high precision in control classification helps maintain clinical efficiency by focusing resources on cases that require intervention. Importance of the Model in Clinical Workflow Integrating the EfficientNet-B0-based model into clinical workflow offers several advantages. Enhanced diagnostic accuracy is achieved by providing precise classification of intra-bony lesions, reducing the likelihood of diagnostic errors, and ensuring that patients receive appropriate and timely treatment. Time efficiency improves as automated classification accelerates the diagnostic process, allowing dental professionals to manage larger patient volumes without compromising accuracy. The model serves as a valuable decision support tool, particularly for less experienced practitioners or in complex cases in which differential diagnosis is challenging. By minimizing the need for manual image interpretation, the model allows dental radiologists to allocate their expertise to more complex diagnostic and treatment-planning tasks, thereby optimizing resource utilization. Ultimately, timely and accurate diagnosis directly correlates with improved patient outcomes, reduced progression of dental pathologies, and enhanced overall oral health. Limitations Despite these promising results, several limitations of this study warrant consideration. The dataset utilized in this study comprised 833 training and validation images, which, while sufficient for initial model training and validation, may not capture the full variability of intra-bony lesions encountered in diverse clinical populations. The relatively small sample size, particularly for classes such as pericoronitis (24 images), may limit the model's generalizability and robustness in real-world settings. This small sample size restricts the implementation of advanced techniques such as objective identification and segmentation algorithms for localized pathology detection 18 – 20 . The retrospective design introduced potential biases related to image selection and quality 21 . Variations in image acquisition protocols, patient positioning, and radiographic equipment across different centers may affect the performance of the model. Additionally, some classes have significantly fewer samples, which can affect the model's ability to effectively learn distinguishing features for those categories. Although data augmentation techniques have been employed, this inherent imbalance remains a challenge. Furthermore, the model was trained and validated using data from specific clinics and centers, limiting the assessment of its performance on external datasets. Prospective validation using independent datasets is necessary to confirm the efficacy of the model across different populations and imaging conditions. Future Directions Future research should focus on expanding the dataset to include a larger and more diverse set of images, encompassing variations in patient demographics, imaging devices, and lesion presentation. This would enhance the ability of the model to generalize across different clinical scenarios and reduce the risk of overfitting. Considering the findings of Yilmaz et al. (2023) 15 and Kurt et al. (2024) 16 , integration Object detection models, such as YOLOv4 22 and YOLOv5 23 , with classification architectures could be explored to develop multifunctional AI tools capable of both developmental assessments and pathology classifications within a single framework. Additionally, incorporating advanced data augmentation techniques, exploring ensemble methods, and leveraging transfer learning from larger pretrained models may further improve classification accuracy and robustness. Prospective studies integrating the model into clinical workflows are necessary to evaluate its practical utility and impact on the diagnostic accuracy and efficiency. Developing user-friendly interfaces and ensuring seamless integration with existing radiological systems would facilitate the adoption of AI-driven tools in routine dental practice. Furthermore, comparative studies assessing the performance of different deep learning architectures across various dental imaging tasks could provide deeper insights into optimizing AI applications for specific diagnostic needs. Conclusion In summary, the EfficientNet-B0-based model developed in this study demonstrates significant potential for accurately classifying intrabony lesions in dental X-ray images. With high validation accuracy and robust F1 scores across multiple classes, this model is a promising tool for enhancing diagnostic precision and efficiency in dental radiology. The comparative analyses of Yilmaz et al. (2023) and Kurt et al. (2024) highlight the diverse applications of deep learning in dentistry, underscoring the versatility and efficacy of AI-driven models in improving dental diagnostics and patient care. Addressing the limitations identified through future research will be pivotal in translating these findings into practical clinical applications, ultimately contributing to improved patient outcomes and advancements in dental health. Declarations Ethics Approval : This study was conducted in accordance with the ethical standards of the institutional research committee of Al Farahidi University/College of Dentistry and with the 1964 Helsinki declaration and its later amendments. Ethical approval was obtained from the Al Farahidi University/College of Dentistry Ethics Committee. Human Ethics and Consent to Participate : All participants provided written informed consent prior to inclusion in the study. The privacy rights of human subjects have been observed. Conflicts of Interest : The authors declare that they have no conflict of interest. Consent for publication: Not Applicable Consent to Participate : Written informed consent was obtained from all participants included in the study. All procedures were performed in accordance with relevant guidelines and regulations. Funding: This research received no external funding Clinical Trial Number: Not applicable. Acknowledgments: We extend our gratitude to Mohamed Ali Bahr for his technical support with the EfficientNet-B0 model implementation and to the staff at Al-Sha'ab Specialized Center for their help with data collection. We appreciate the constructive comments from our peer reviewers. Advances in knowledge : This study presents the first application of EfficientNet-B0 architecture for the comprehensive classification of eight distinct intrabony lesion types in dental radiographs, achieving superior accuracy compared to existing approaches while maintaining computational efficiency. References Shaikh MS, Zafar MS, Alnazzawi A, Javed F. Nanocrystalline hydroxyapatite in regeneration of periodontal intrabony defects: A systematic review and meta-analysis. Annals Anatomy - Anatomischer Anzeiger. 2022;240:151877. 10.1016/j.aanat.2021.151877 . Suzuki K. Overview of deep learning in medical imaging. Radiol Phys Technol. 2017;10(3):257–73. 10.1007/s12194-017-0406-5 . Farias JG, Souza RCA, Hassam SF, Cardoso JA, Ramos TCF, Santos HKA. Epidemiological study of intraosseous lesions of the stomatognathic or maxillomandibular complex diagnosed by a Reference Centre in Brazil from 2006–2017. Br J Oral Maxillofac Surg. 2019;57(7):632–7. 10.1016/j.bjoms.2019.05.003 . Webster S, Fraser J. Artificial intelligence and dental panoramic radiographs: where are we now? Evid Based Dent. 2024;25(1):43–4. 10.1038/s41432-024-00978-9 . Hingst V, Weber MA. [Dental X-ray diagnostics with the orthopantomography - Technique and typical imaging results]. Radiologe. 2020;60(1):77–92. 10.1007/s00117-019-00620-1 . Lv Q, Zhang S, Wang Y. Deep Learning Model of Image Classification Using Machine Learning. Li Q, ed. Advances in Multimedia . 2022;2022:1–12. 10.1155/2022/3351256 Du X, Chen Y, Zhao J, Xi Y. A Convolutional Neural Network Based Auto-Positioning Method For Dental Arch In Rotational Panoramic Radiography. Annu Int Conf IEEE Eng Med Biol Soc. 2018;2018:2615–8. 10.1109/EMBC.2018.8512732 . Kansal K, Chandra TB, Singh A. ResNet-50 vs. EfficientNet-B0: Multi-Centric Classification of Various Lung Abnormalities Using Deep Learning. Procedia Comput Sci. 2024;235:70–80. 10.1016/j.procs.2024.04.007 . Hwang JJ, Jung YH, Cho BH, Heo MS. An overview of deep learning in the field of dentistry. Imaging Sci Dent. 2019;49(1):1–7. 10.5624/isd.2019.49.1.1 . Hasnain MA, Ali Z, Maqbool MS, Aziz M. X-ray Image Analysis for Dental Disease: A Deep Learning Approach Using EfficientNets. VFAST trans softw eng. 2024;12(3):147–65. 10.21015/vtse.v12i3.1912 . Lin S, Hao X, Liu Y, Yan D, Liu J, Zhong M. Lightweight deep learning methods for panoramic dental X-ray image segmentation. Neural Comput Applic. 2023;35(11):8295–306. 10.1007/s00521-022-08102-7 . Ali MA, Fujita D, Kobashi S. Teeth and prostheses detection in dental panoramic X-rays using CNN-based object detector and a priori knowledge-based algorithm. Sci Rep. 2023;13(1):16542. 10.1038/s41598-023-43591-z . albumentations — albumentations 1.1.0 documentation. Accessed November 10. 2024. https://albumentations.readthedocs.io/en/latest/index.html CrossEntropyLoss. — PyTorch 2.5 documentation. Accessed November 10, 2024. https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html Yilmaz S, Tasyurek M, Amuk M, Celik M, Canger EM. Developing deep learning methods for classification of teeth in dental panoramic radiography. Oral Surg Oral Med Oral Pathol Oral Radiol. 2024;138(1):118–27. 10.1016/j.oooo.2023.02.021 . Kurt A, Günaçar DN, Şılbır FY, et al. Evaluation of tooth development stages with deep learning-based artificial intelligence algorithm. BMC Oral Health. 2024;24(1):1034. 10.1186/s12903-024-04786-6 . Lee JH, Kim DH, Jeong SN. Diagnosis of cystic lesions using panoramic and cone beam computed tomographic images based on deep learning neural network. Oral Dis. 2020;26(1):152–8. 10.1111/odi.13223 . Leite AF, Gerven AV, Willems H, et al. Artificial intelligence-driven novel tool for tooth detection and segmentation on panoramic radiographs. Clin Oral Investig. 2021;25(4):2257–67. 10.1007/s00784-020-03544-6 . Muramatsu C, Morishita T, Takahashi R, et al. Tooth detection and classification on panoramic radiographs for automatic dental chart filing: improved classification by multi-sized input data. Oral Radiol. 2021;37(1):13–9. 10.1007/s11282-019-00418-w . Lee JH, Han SS, Kim YH, Lee C, Kim I. Application of a fully deep convolutional neural network to the automation of tooth segmentation on panoramic radiographs. Oral Surg Oral Med Oral Pathol Oral Radiol. 2020;129(6):635–42. 10.1016/j.oooo.2019.11.007 . Weese J, Lorenz C. Four challenges in medical image analysis from an industrial perspective. Med Image Anal. 2016;33:44–9. 10.1016/j.media.2016.06.023 . Li P, Han T, Ren Y, Xu P, Yu H. Improved YOLOv4-tiny based on attention mechanism for skin detection. PeerJ Comput Sci. 2023;9:e1288. 10.7717/peerj-cs.1288 . Zhang J, Zhang J, Zhou K, Zhang Y, Chen H, Yan X. An Improved YOLOv5-Based Underwater Object-Detection Framework. Sens (Basel). 2023;23(7):3693. 10.3390/s23073693 . Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Editorial decision: Revision requested 29 Jul, 2025 Reviews received at journal 20 Jul, 2025 Reviewers agreed at journal 03 Jul, 2025 Reviews received at journal 12 May, 2025 Reviewers agreed at journal 07 May, 2025 Reviewers invited by journal 02 Apr, 2025 Editor assigned by journal 01 Apr, 2025 Submission checks completed at journal 01 Apr, 2025 First submitted to journal 01 Apr, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6207574","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":445268633,"identity":"868b2df2-b17d-4601-9edb-c00a6d2de26d","order_by":0,"name":"Ammar Mohi","email":"","orcid":"","institution":"BDS, Al-Farahidi University","correspondingAuthor":false,"prefix":"","firstName":"Ammar","middleName":"","lastName":"Mohi","suffix":""},{"id":445268634,"identity":"e2ecdfb4-cfe2-447b-a2ec-e637e63cc392","order_by":1,"name":"Lubna Kareem","email":"","orcid":"","institution":"BDS, HDD, Al-Sha'ab Specialized Dental Center","correspondingAuthor":false,"prefix":"","firstName":"Lubna","middleName":"","lastName":"Kareem","suffix":""},{"id":445268635,"identity":"c097fbe0-3bd3-4117-a5e0-15ac5c0556ce","order_by":2,"name":"Atef A Hassan","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABBUlEQVRIiWNgGAWjYJCCD4wNzAz87O0HH3yoAHKZmRvwq2djYJwB0iLZcybZcMYZkBZGIrUY3HAwk+ZtAwkR0MI/v/lgw88d1nIMNxjSpHnn1UbztwO1/KjYhlOLxDG2xMbeM+nGjLMbD1vO3XY8d8ZhxgbGnjO3cVtzjMf8AW/b4cRmmQOJN95uO5bbANTCzNiGW4v8Mf6PjX+BWtokEgwkeOccy51PSIvBMR7GZpAtPRIJRpK8DTW5GwhpMTyWZtgsC/SLBA8okI8dyN0I1HIQn1/kDh9+2PgWGGL2x0FRWVOXO+/84YMPflTg8T4aOAwmDxCtHgjqSFE8CkbBKBgFIwQAAHluZbqyZ2SLAAAAAElFTkSuQmCC","orcid":"","institution":"Al-Azhar University","correspondingAuthor":true,"prefix":"","firstName":"Atef","middleName":"A","lastName":"Hassan","suffix":""}],"badges":[],"createdAt":"2025-03-12 01:38:05","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6207574/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6207574/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":81931080,"identity":"ee7aa485-1029-46aa-91bc-b97dbce4e64b","added_by":"auto","created_at":"2025-05-05 05:28:10","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":36865,"visible":true,"origin":"","legend":"\u003cp\u003eClass distribution of different classes\u003c/p\u003e","description":"","filename":"1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/dd82e3e22684b288e2a242a4.jpg"},{"id":81931102,"identity":"3a3e7429-683f-4515-80e9-d1a15fadaae7","added_by":"auto","created_at":"2025-05-05 05:28:14","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":61923,"visible":true,"origin":"","legend":"\u003cp\u003eProgression of Training and Validation Accuracy Over 50 Epochs\u003c/p\u003e","description":"","filename":"2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/c29392c7f4c938c8e414961e.jpg"},{"id":81931081,"identity":"6ea2da36-37b5-470a-83b3-8da86dae70d7","added_by":"auto","created_at":"2025-05-05 05:28:10","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":62111,"visible":true,"origin":"","legend":"\u003cp\u003eDecline in Training and Validation Loss Over 50 Epochs\u003c/p\u003e","description":"","filename":"3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/bee232cdf13a5f277b460045.jpg"},{"id":81932447,"identity":"232e826f-caff-4a9c-8a98-483012aa7bb7","added_by":"auto","created_at":"2025-05-05 05:36:10","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":67711,"visible":true,"origin":"","legend":"\u003cp\u003eTrends in Precision, Recall, and F1-Score Across Training Epochs\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/ad9869d0e98d1496c483f4ca.jpg"},{"id":81932449,"identity":"00105f72-2de7-4e3f-9161-d63ef622af60","added_by":"auto","created_at":"2025-05-05 05:36:10","extension":"jpg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":58950,"visible":true,"origin":"","legend":"\u003cp\u003eNormalized Confusion Matrix of the Model's Predictions\u003c/p\u003e","description":"","filename":"5.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/8de8503799abe1884f66807e.jpg"},{"id":81931082,"identity":"1b1be839-8d2a-43b3-98f8-5c68c49a98c1","added_by":"auto","created_at":"2025-05-05 05:28:10","extension":"jpg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":120306,"visible":true,"origin":"","legend":"\u003cp\u003eshows examples of misclassified intrabony lesions by the EfficientNet-B0 model\u003c/p\u003e","description":"","filename":"6.jpg","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/b82394d4099f6bf627886619.jpg"},{"id":81935403,"identity":"1b828db5-157a-4214-b5ff-010105605fb2","added_by":"auto","created_at":"2025-05-05 06:00:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1248206,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6207574/v1/709faf01-ed85-42be-8485-5b6d853b9cf7.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"EfficientNet-B0 Deep Learning Model for Accurate Classification of Intrabony Lesions in Dental Panoramic Radiographs","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAccurate diagnosis of intrabony lesions is pivotal in dental practice, as these conditions can significantly impact oral health and overall patient wellbeing \u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e,\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Intrabony lesions, including periapical abscesses, condensing osteitis, and radicular cysts, often result from chronic infections, trauma, or developmental anomalies. Timely and precise identification of these lesions through radiographic imaging is essential for determining appropriate treatment strategies, preventing disease progression, and enhance patient outcomes\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eOrthopantomograms (OPGs), or panoramic radiographs, are widely used in dental diagnostics because of their ability to provide a comprehensive view of the maxillofacial region in a single image\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. However, the interpretation of OPGs requires substantial expertise, and the manual analysis process is subject to inter-observer variability and potential diagnostic errors\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. These challenges underscore the need for automated, reliable, and efficient diagnostic tools to assist dental professionals in accurately classifying intrabony lesions.\u003c/p\u003e \u003cp\u003eIn recent years, deep learning (DL) and convolutional neural networks (CNNs) have emerged as transformative technologies in medical imaging, offering unprecedented accuracy and efficiency in image classification and pattern recognition tasks\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e,\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u003c/sup\u003e. Among the various CNN architectures, EfficientNet-B0 has gained prominence for its compound scaling method, which uniformly scales the network depth, width, and resolution, resulting in superior performance with fewer parameters compared to other models\u003csup\u003e\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e,\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. This efficiency makes EfficientNet-B0 particularly suitable for applications that require high-resolution image analysis such as dental radiology\u003csup\u003e\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eDespite advancements in DL applications in medical imaging, there remains a gap in the automated classification of diverse intrabony lesions on dental radiographs. Most existing studies have focused on specific conditions or have utilized architectures primarily designed for object detection rather than comprehensive image classification\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e,\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. To address this gap, the present study aimed to develop and evaluate an EfficientNet-B0-based model for the accurate classification of intrabony lesions in dental X-ray images. By leveraging a robust dataset and advanced preprocessing techniques, this study sought to enhance diagnostic precision and support clinical decision-making in dental radiology.\u003c/p\u003e"},{"header":"Methods and materials/patients","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy Design\u003c/h2\u003e \u003cp\u003e This retrospective study was meticulously designed in accordance with established dental protocols and ethical standards across multiple private dental clinics and specialized centers. Ethical approval was obtained from the [Al Farahidi University / Collage of Dentistry] to ensure compliance with all regulatory guidelines. Oversight was provided by Two Oral and Maxillofacial (OMF) radiologists to maintain adherence to diagnostic criteria and ethical protocols. The primary objective was to develop and evaluate an EfficientNet-B0-based convolutional neural network (CNN) model for the accurate classification of intrabony lesions in dental X-ray images by leveraging advanced deep learning techniques to enhance diagnostic precision and support clinical decision-making.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eEligibility Criteria\u003c/h3\u003e\n\u003cp\u003eOrthopantomogram (OPG) panoramic radiographs were collected between October 2021 and April 2023 at the Basamat Private Dental Clinic and Al-Sha'ab Specialized Center. The inclusion criteria were age between 18 and 70 years, absence of previous jaw trauma or fractures, presence of partial or complete edentulous alveolar bone, broad dental inclusion dimensions within the included dew (FOV) of the OPG image, high-resolution images devoid of artifacts, and achieved centric occlusion during OPG exposure with proper head positioning. Exclusion criteria eliminated Patients with a history of facial trauma, dentofacial malformations or anomaly syndromes, developmental bone diseases, osteometabolic disorders, or hormonal diseases affecting bone morphology were excluded.\u003c/p\u003e\n\u003ch3\u003eStudy Sample\u003c/h3\u003e\n\u003cp\u003eA total of 245 patients' OPG images were meticulously screened and included in the analysis, resulting in 833 training and validation samples categorized into eight distinct classes: Periapical Widening, Condensing Osteitis, Periapical Granuloma, Nil Control (no pathological findings), Diffuse Lesion, Periapical Abscess, Pericoronitis, and Radicular Cyst. Each class represents a specific intrabony lesion selected based on its clinical significance and frequency of occurrence in dental radiology.\u003c/p\u003e\n\u003ch3\u003eData Collection\u003c/h3\u003e\n\u003cp\u003eRadiographic imaging was performed using a Myray 2D Pro X9 Panoramic Device (FOV 15 \u0026times; 15 cm; 50\u0026ndash;100 kVp; 1\u0026ndash;22 mA; 20 \u0026micro;s) from Myray (Cephla, Bologna, Italy). This device was chosen for its high-resolution imaging capabilities, which are essential for detailed evaluation of intra-bony bone morphology, density, and spatial positioning based on anatomical and radiological principles. Each intra-alveolar bone pathology was categorized as radiopaque or radiolucent and further classified according to site, position, and size.\u003c/p\u003e\n\u003ch3\u003eData Preprocessing and Augmentation\u003c/h3\u003e\n\u003cp\u003eData preprocessing and augmentation were performed using a custom PyTorch Dataset class, the DentalXRayDataset, designed to validate directory structures, retrieve class names, and compile lists of image file paths with corresponding labels. Data augmentation techniques were employed using the Albumentation library\u003csup\u003e\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e to enhance the model's generalization capabilities and simulate real-world variability inherent in clinical radiographs. The training data were resized to 224 \u0026times; 224 pixels, with random rotations within \u0026plusmn;\u0026thinsp;20 \u0026deg;, horizontal flipping with a training ability of 0.5, brightness and contrast adjustments with a probability of 0.2, and normalization on ImageNet statistics. Advanced augmentations, including random 90-degree rotations, Gaussian noise, Gaussian blur, motion blur, median blur, and general blurring, were incorporated to further diversify the training data. To maintain the integrity of the evaluation metrics, the validation data underwent minimal preprocessing, limited to resizing and normalization.\u003c/p\u003e \u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eHandling Class Imbalance\u003c/h2\u003e \u003cp\u003eAddressing the inherent class imbalance, particularly with classes such as pericoronitis (24 images), involves computing class weights using the compute_class_weight function from scikit-learn. These weights were integrated into the CrossEntropyLoss function\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e, allowing the model to assign appropriate emphasis to minority classes during training. Additionally, aggressive data augmentation techniques have been applied to augment minority classes, thereby increasing their effective sample size and diversity.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eModel Architecture\u003c/h3\u003e\n\u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003eEfficientNet-B0 Configuration\u003c/h2\u003e \u003cp\u003eThe classification model was built using the EfficientNet-B0 architecture, selected for its compound scaling method, which uniformly scales the network depth, width, and resolution, achieving superior performance with fewer parameters. The architecture was modified to accommodate the specific requirements of the study by configuring the input layer to accept images with a size of 224 \u0026times; 224 pixels, replacing the final fully connected layer to match the number of target classes (eight), and incorporating a dropout layer to prevent overfitting.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eOptimizer and Scheduler\u003c/h2\u003e \u003cp\u003eTraining was conducted over 50 epochs using the Adam optimizer with an initial learning rate of 0.001 and weight decay of 0.0005 to prevent overfitting by penalizing large weights. A StepLR scheduler was utilized, reducing the learning rate by a factor of 0.1 every three epochs to facilitate finer weight adjustments as training progressed, promoting stabilization, and preventing overfitting.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eMixed Precision Training and Gradient Scaling\u003c/h2\u003e \u003cp\u003eTo optimize the training efficiency and leverage the computational capabilities of modern GPUs, mixed-precision training was implemented using Torch. cuda. amp. This approach, coupled with gradient scaling via GradScaler, accelerates the training process and reduces memory consumption without compromising the model performance.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003eTraining Procedure\u003c/h2\u003e \u003cp\u003eThe training loop involves a systematic process in which each epoch includes a forward pass through the EfficientNet-B0 model to obtain predictions, computation of cross-entropy loss incorporating class weights, backpropagation using mixed-precision training with autocast, and optimizer updates based on scaled gradients. The learning rate scheduler adjusts the learning rate based on a predefined schedule. After each epoch, performance metrics, such as accuracy, precision, recall, F1-score, and confusion matrix, were calculated for the validation set. The state of the model was saved whenever an improvement in the validation accuracy was observed, ensuring the retention of the best-performing model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec14\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation Metrics\u003c/h2\u003e \u003cp\u003eThe model performance was rigorously evaluated using a suite of metrics to ensure a holistic assessment of classification capabilities. The metrics included the accuracy, precision, recall, F1-score, confusion matrix, and specificity. Precision measures the ratio of true positives to the sum of true and false positives, whereas recall (sensitivity) assesses the ratio of true positives to the sum of true positives and false negatives. The F1-score provided a balance between precision and recall. The confusion matrix offered a detailed breakdown of true versus predicted classifications across all classes, and the specificity evaluated the model's ability to correctly identify non-pathological cases.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cp\u003eThe dental X-ray classification model built upon the EfficientNet-B0 architecture demonstrated exceptional performance in classifying intrabony lesions across eight distinct categories. The model was trained and validated using a dataset of 833 images with Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e showing the class distributions of different classes in the database.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eOverall Evaluation Metrics\u003c/h2\u003e \u003cp\u003eAs shown in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e, the model achieved a final validation accuracy of 93.04% and a validation loss of 0.2135. In addition, the model demonstrated robust performance across various metrics, including the precision, recall, and F1-score, with values of 0.9345, 0.9304, and 0.9295, respectively.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eFinal Evaluation Metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMetric\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eValue\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal Validation Accuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e93.04%\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal Validation Loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.2135\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal Precision\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9345\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal Recall\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9304\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eFinal F1 Score\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9295\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"2\"\u003eClass-wise Performance\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe model's precision, recall, and F1-score metrics underscore its robust performance across all categories. Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e details the class-wise metrics, highlighting the high precision and recall values, which indicate the model's proficiency in correctly identifying true positives, while maintaining low rates of false positives and negatives.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClass-wise Precision, Recall, and F1-Score\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"4\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eClass\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eF1-Score\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeriapical Widening\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.87\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCondensing Osteitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeriapical Granuloma\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.90\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.89\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNil Control\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiffuse Lesion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.88\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.85\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.86\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePeriapical Abscess\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePericoronitis\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.93\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.91\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.92\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRadicular Cyst\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.95\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.94\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"4\"\u003eTraining Dynamics\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eTable\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e presents a snapshot of the model performance across the selected epochs, illustrating a progressive improvement over 50 epochs. Significant enhancements were evident from epochs 1 to 20, when the model rapidly increased in accuracy and decreased in loss. By epoch 50, the model achieved high validation accuracy with minimal loss, signifying effective learning and generalization.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eEpoch-wise Training and Validation Metrics\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEpoch\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTrain Loss\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eTrain Acc (%)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eVal Loss\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eVal Acc (%)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.0224\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e23.89\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e2.2062\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e19.81\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e10\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9202\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e69.27\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9202\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e69.27\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e20\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.2130\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e93.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2130\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e93.04\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e30\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.2528\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.24\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2528\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e91.24\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e40\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.2130\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e93.04\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2130\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e93.04\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003e50\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.1998\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e91.48\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.2135\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e93.04\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe accuracy plots demonstrated a consistent upward trajectory for both training and validation accuracies. The early epochs exhibited rapid improvements, which gradually stabilized as the model approached convergence. The final training and validation accuracies reached 91.48% and 93.04%, respectively, indicating effective learning and generalization.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe loss curves exhibited a steady decline in both training and validation losses. The training loss decreases from an initial value of 2.0224 to a final value of 0.1998, whereas the validation loss decreases from 2.2062 to 0.2135. This consistent reduction indicates an improved capacity of the model to minimize prediction errors.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe precision, recall, and F1-score plots indicate steady improvement throughout the training process. By epoch 50, the model achieved high-precision recall values, resulting in an F1-score of 0.9295. This convergence suggests that the model effectively balances precision and recall, minimizing both false positives and negatives.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe confusion matrix provides a detailed breakdown of the classification performance of the model across all classes. High values along the diagonal and low off-diagonal values demonstrated the model's ability to accurately distinguish between different dental conditions with minimal misclassifications.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe learning rate scheduler played a pivotal role in the training process of the model. The learning rate was reduced at predefined intervals (every three epochs). The scheduler facilitates finer weight updates, contributing to the stabilization of training and the prevention of overfitting. This strategic adjustment enhances the model\u0026rsquo;s ability to converge to an optimal solution.\u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis study successfully developed and evaluated an EfficientNet-B0-based model for the classification of intrabony lesions in dental X-ray images. Achieving a validation accuracy of 93.04% and a robust F1 score of 0.9295, the model demonstrates strong performance across eight distinct categories, including Periapical Widening, Condensing Osteitis, Periapical Granuloma, Nil Control, Diffuse Lesion, Periapical Abscess, Pericoronitis, and Radicular Cyst. These results underscore the potential of EfficientNet-B0 to enhance the diagnostic accuracy and efficiency in dental radiology.\u003c/p\u003e \u003cdiv id=\"Sec18\" class=\"Section2\"\u003e \u003ch2\u003eInterpretation of Results\u003c/h2\u003e \u003cp\u003eThe high validation accuracy and F1 score indicated that the EfficientNet-B0 model effectively distinguished between various intra-bony lesions with minimal misclassifications. Notably, the model exhibited superior performance in classes such as Nil Control (precision, 0.95; recall, 0.94) and radicular cysts (precision, 0.94; recall, 0.95), suggesting its proficiency in identifying conditions with distinct radiographic features. Conversely, classes such as periapical widening and diffuse lesions recorded slightly lower precision and recall values (0.87 and 0.88, respectively), which may be attributed to overlapping radiographic characteristics with other lesion types or inherent variability within these conditions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec19\" class=\"Section2\"\u003e \u003ch2\u003eComparative Analysis with Recent Studies\u003c/h2\u003e \u003cp\u003eA recent study by Yilmaz et al. (2023)\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e employed YOLOv4 and Faster R-CNN, which are primarily designed for object detection tasks and can localize and classify objects within an image. This study focused on tooth classification, identifying 36 classes, including 32 teeth and four impacted teeth, in a dataset of 1,200 panoramic radiographs. The YOLOv4 model achieved outstanding precision (99.90%), recall (99.18%), and F1 score (99.54%), thereby significantly outperforming the Faster R-CNN method. In contrast, our study utilized EfficientNet-B0, an image classification model optimized for categorizing entire images rather than detecting and localizing individual objects. In addition, Yilmaz et al\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. utilized a larger dataset of 1,200 images with 36 classes, focusing on individual tooth identification and classification. However, our study used 833 training and validation images across eight intra-bony lesion classes, representing a more complex classification task owing to the nuanced differences between the lesion types. The task of Yilmaz et al\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. involves the identification and classification of individual teeth, which generally present more distinct and localized features. In contrast, our task involves classifying various types of intra-bony lesions that can exhibit overlapping and subtle radiographic characteristics, increasing the complexity of the classification process.\u003c/p\u003e \u003cp\u003eWhile Yilmaz et al\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. achieved near-perfect precision and recall in tooth classification tasks, our EfficientNet-B0 model demonstrated strong performance in a more complex classification scenario involving varied lesion types, achieving a precision of 0.9345, recall of 0.9304, and an F1-score of 0.9295. The slight reduction in performance metrics is attributable to the inherent complexity of distinguishing intra-bony lesions, which often requires discerning subtle differences in bone morphology and density, which are less prominent in tooth identification.\u003c/p\u003e \u003cp\u003eKurt et al\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. utilized YOLOv5, which is designed for real-time object detection and classification, enabling the simultaneous localization and classification of objects within an image. Our study employed EfficientNet-B0, which was optimized for image-classification tasks without explicit localization capabilities. Additionally, Kurt et al. analyzed 1,500 images with 46 classes corresponding to different tooth development stages and compared them to our 833 training and validation images across eight lesion categories. Kurt et al. focused on detecting and classifying tooth development stages, a task that benefits from the precise localization and high sensitivity offered by YOLOv5. In contrast, our study involved classifying intrabony lesions, requiring a model to differentiate between multiple lesion types with overlapping features.\u003c/p\u003e \u003cp\u003eKurt et al\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. reported a sensitivity of 0.99, precision of 0.72, and an F1 score of 0.84. A high sensitivity indicates excellent true-positive detection, which is crucial for ensuring that the most relevant developmental stages are identified. However, the lower precision suggests a higher rate of false positives compared to our model.\u003c/p\u003e \u003cp\u003eOur EfficientNet-B0 model achieved a balanced precision (0.9345) and recall (0.9304), culminating in a robust F1 score (0.9295). This balance minimizes both false positives and false negatives, and enhances the reliability of diagnoses in clinical settings. While YOLOv5 excels in tasks requiring high sensitivity, such as developmental stage detection, EfficientNet-B0 offers a balanced approach that is ideal for complex lesion classification, where both precision and recall are paramount.\u003c/p\u003e \u003cp\u003eOur study was part of a larger project that aimed to collect additional data and images covering a wider range of classes and pathologies. The project will implement more advanced techniques and tools to assist specialists in saving time and effort by developing a web-based software application built on this model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec20\" class=\"Section2\"\u003e \u003ch2\u003eModel-Specific Insights\u003c/h2\u003e \u003cdiv id=\"Sec21\" class=\"Section3\"\u003e \u003ch2\u003eEfficientNet-B0 Architecture\u003c/h2\u003e \u003cp\u003eEfficientNet-B0 is renowned for its compound scaling method, which uniformly scales all dimensions of depth, width, and resolution using a simple yet highly effective compound coefficient. This balanced scaling enables EfficientNet-B0 to achieve superior performance with fewer parameters than the other models, making it both computationally efficient and highly accurate. In dental radiology, where high-resolution images are paramount for an accurate diagnosis, EfficientNet-B0's ability to extract intricate features from detailed radiographs is particularly advantageous.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec22\" class=\"Section2\"\u003e \u003ch2\u003eFeature Extraction and Learning Efficiency\u003c/h2\u003e \u003cp\u003eThe architecture of the EfficientNet-B0 model facilitates effective feature extraction through its deep and narrow layers, allowing it to capture both global and local patterns within radiographic images. This capability is crucial for differentiating various intrabony lesions that may exhibit subtle differences in bone density, morphology, and spatial positioning. Additionally, the model's efficient use of computational resources ensures faster training and inference times, making it suitable for real-time clinical applications in which timely decision making is essential.\u003c/p\u003e \u003cdiv id=\"Sec23\" class=\"Section3\"\u003e \u003ch2\u003eData Augmentation and Preprocessing\u003c/h2\u003e \u003cp\u003eThe implementation of data augmentation techniques such as rotation, horizontal flipping, and brightness adjustments plays a pivotal role in enhancing the generalization capabilities of the model. These augmentations simulate the variability inherent in clinical radiographs, such as differences in patient positioning and imaging conditions, thereby reducing the risk of overfitting and improving the model robustness across diverse clinical scenarios.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e \u003cdiv id=\"Sec24\" class=\"Section2\"\u003e \u003ch2\u003eClinical Importance in Management of Different Conditions\u003c/h2\u003e \u003cp\u003eThe ability to accurately classify intrabony lesions has profound implications for the clinical management of dentistry. Each lesion type identified by the model corresponded to a distinct pathological condition that required specific treatment approaches.\u003c/p\u003e \u003cp\u003ePeriapical Widening and Periapical Abscess are conditions that often result from chronic pulpitis and can lead to significant bone loss and tooth mobility if left untreated. Early and accurate detection enables timely root canal therapy or extraction, preventing further complications, such as bone destruction and systemic infections.\u003c/p\u003e \u003cp\u003eCondensing osteitis and radicular cysts present different clinical challenges. Condensing osteitis is typically a response to low-grade inflammation, whereas radicular cysts are developmental lesions that arise from epithelial remnants. Differentiating between these conditions is crucial, as condensing osteitis may resolve with conservative treatment, whereas radicular cysts might require surgical intervention to prevent bone erosion and maintain jaw integrity.\u003c/p\u003e \u003cp\u003ePericoronitis involves inflammation around partially erupted molars, often leading to pain and infection, whereas Diffuse Lesions can indicate a variety of pathologies, including benign and malignant conditions\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. Accurate classification assists in determining the appropriate intervention, whether it is antimicrobial therapy, removal of the offending tooth in pericoronitis, or further diagnostic imaging and biopsy for diffuse lesions.\u003c/p\u003e \u003cp\u003eNil Control classifications are equally important, as correctly identifying images with no pathological findings is essential to avoid unnecessary treatment and alleviate patient anxiety. Ensuring high precision in control classification helps maintain clinical efficiency by focusing resources on cases that require intervention.\u003c/p\u003e \u003cdiv id=\"Sec25\" class=\"Section3\"\u003e \u003ch2\u003eImportance of the Model in Clinical Workflow\u003c/h2\u003e \u003cp\u003eIntegrating the EfficientNet-B0-based model into clinical workflow offers several advantages. Enhanced diagnostic accuracy is achieved by providing precise classification of intra-bony lesions, reducing the likelihood of diagnostic errors, and ensuring that patients receive appropriate and timely treatment. Time efficiency improves as automated classification accelerates the diagnostic process, allowing dental professionals to manage larger patient volumes without compromising accuracy.\u003c/p\u003e \u003cp\u003eThe model serves as a valuable decision support tool, particularly for less experienced practitioners or in complex cases in which differential diagnosis is challenging. By minimizing the need for manual image interpretation, the model allows dental radiologists to allocate their expertise to more complex diagnostic and treatment-planning tasks, thereby optimizing resource utilization. Ultimately, timely and accurate diagnosis directly correlates with improved patient outcomes, reduced progression of dental pathologies, and enhanced overall oral health.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec26\" class=\"Section3\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eDespite these promising results, several limitations of this study warrant consideration. The dataset utilized in this study comprised 833 training and validation images, which, while sufficient for initial model training and validation, may not capture the full variability of intra-bony lesions encountered in diverse clinical populations. The relatively small sample size, particularly for classes such as pericoronitis (24 images), may limit the model's generalizability and robustness in real-world settings. This small sample size restricts the implementation of advanced techniques such as objective identification and segmentation algorithms for localized pathology detection\u003csup\u003e\u003cspan additionalcitationids=\"CR19\" citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe retrospective design introduced potential biases related to image selection and quality\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Variations in image acquisition protocols, patient positioning, and radiographic equipment across different centers may affect the performance of the model. Additionally, some classes have significantly fewer samples, which can affect the model's ability to effectively learn distinguishing features for those categories. Although data augmentation techniques have been employed, this inherent imbalance remains a challenge.\u003c/p\u003e \u003cp\u003eFurthermore, the model was trained and validated using data from specific clinics and centers, limiting the assessment of its performance on external datasets. Prospective validation using independent datasets is necessary to confirm the efficacy of the model across different populations and imaging conditions.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec27\" class=\"Section3\"\u003e \u003ch2\u003eFuture Directions\u003c/h2\u003e \u003cp\u003eFuture research should focus on expanding the dataset to include a larger and more diverse set of images, encompassing variations in patient demographics, imaging devices, and lesion presentation. This would enhance the ability of the model to generalize across different clinical scenarios and reduce the risk of overfitting.\u003c/p\u003e \u003cp\u003eConsidering the findings of Yilmaz et al. (2023)\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e and Kurt et al. (2024)\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e, integration Object detection models, such as YOLOv4\u003csup\u003e22\u003c/sup\u003e and YOLOv5\u003csup\u003e23\u003c/sup\u003e, with classification architectures could be explored to develop multifunctional AI tools capable of both developmental assessments and pathology classifications within a single framework. Additionally, incorporating advanced data augmentation techniques, exploring ensemble methods, and leveraging transfer learning from larger pretrained models may further improve classification accuracy and robustness.\u003c/p\u003e \u003cp\u003eProspective studies integrating the model into clinical workflows are necessary to evaluate its practical utility and impact on the diagnostic accuracy and efficiency. Developing user-friendly interfaces and ensuring seamless integration with existing radiological systems would facilitate the adoption of AI-driven tools in routine dental practice. Furthermore, comparative studies assessing the performance of different deep learning architectures across various dental imaging tasks could provide deeper insights into optimizing AI applications for specific diagnostic needs.\u003c/p\u003e \u003c/div\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn summary, the EfficientNet-B0-based model developed in this study demonstrates significant potential for accurately classifying intrabony lesions in dental X-ray images. With high validation accuracy and robust F1 scores across multiple classes, this model is a promising tool for enhancing diagnostic precision and efficiency in dental radiology. The comparative analyses of Yilmaz et al. (2023) and Kurt et al. (2024) highlight the diverse applications of deep learning in dentistry, underscoring the versatility and efficacy of AI-driven models in improving dental diagnostics and patient care. Addressing the limitations identified through future research will be pivotal in translating these findings into practical clinical applications, ultimately contributing to improved patient outcomes and advancements in dental health.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics Approval\u003c/strong\u003e: This study was conducted in accordance with the ethical standards of the institutional research committee of Al Farahidi University/College of Dentistry and with the 1964 Helsinki declaration and its later amendments. Ethical approval was obtained from the Al Farahidi University/College of Dentistry Ethics Committee.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eHuman Ethics and Consent to Participate\u003c/strong\u003e: All participants provided written informed consent prior to inclusion in the study. The privacy rights of human subjects have been observed.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConflicts of Interest\u003c/strong\u003e: The authors declare that they have no conflict of interest.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u003c/strong\u003e \u003cstrong\u003eNot Applicable\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent to Participate\u003c/strong\u003e: Written informed consent was obtained from all participants included in the study. All procedures were performed in accordance with relevant guidelines and regulations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u0026nbsp;\u003c/strong\u003eThis research received no external funding\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eClinical Trial Number: Not applicable.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments:\u0026nbsp;\u003c/strong\u003eWe extend our gratitude to Mohamed Ali Bahr for his technical support with the EfficientNet-B0 model implementation and to the staff at Al-Sha\u0026apos;ab Specialized Center for their help with data collection. We appreciate the constructive comments from our peer reviewers.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAdvances in knowledge\u003c/strong\u003e: This study presents the first application of EfficientNet-B0 architecture for the comprehensive classification of eight distinct intrabony lesion types in dental radiographs, achieving superior accuracy compared to existing approaches while maintaining computational efficiency.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eShaikh MS, Zafar MS, Alnazzawi A, Javed F. Nanocrystalline hydroxyapatite in regeneration of periodontal intrabony defects: A systematic review and meta-analysis. Annals Anatomy - Anatomischer Anzeiger. 2022;240:151877. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.aanat.2021.151877\u003c/span\u003e\u003cspan address=\"10.1016/j.aanat.2021.151877\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSuzuki K. Overview of deep learning in medical imaging. Radiol Phys Technol. 2017;10(3):257\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s12194-017-0406-5\u003c/span\u003e\u003cspan address=\"10.1007/s12194-017-0406-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFarias JG, Souza RCA, Hassam SF, Cardoso JA, Ramos TCF, Santos HKA. Epidemiological study of intraosseous lesions of the stomatognathic or maxillomandibular complex diagnosed by a Reference Centre in Brazil from 2006\u0026ndash;2017. Br J Oral Maxillofac Surg. 2019;57(7):632\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.bjoms.2019.05.003\u003c/span\u003e\u003cspan address=\"10.1016/j.bjoms.2019.05.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWebster S, Fraser J. Artificial intelligence and dental panoramic radiographs: where are we now? Evid Based Dent. 2024;25(1):43\u0026ndash;4. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41432-024-00978-9\u003c/span\u003e\u003cspan address=\"10.1038/s41432-024-00978-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHingst V, Weber MA. [Dental X-ray diagnostics with the orthopantomography - Technique and typical imaging results]. Radiologe. 2020;60(1):77\u0026ndash;92. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00117-019-00620-1\u003c/span\u003e\u003cspan address=\"10.1007/s00117-019-00620-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLv Q, Zhang S, Wang Y. Deep Learning Model of Image Classification Using Machine Learning. Li Q, ed. \u003cem\u003eAdvances in Multimedia\u003c/em\u003e. 2022;2022:1\u0026ndash;12. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1155/2022/3351256\u003c/span\u003e\u003cspan address=\"10.1155/2022/3351256\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDu X, Chen Y, Zhao J, Xi Y. A Convolutional Neural Network Based Auto-Positioning Method For Dental Arch In Rotational Panoramic Radiography. Annu Int Conf IEEE Eng Med Biol Soc. 2018;2018:2615\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1109/EMBC.2018.8512732\u003c/span\u003e\u003cspan address=\"10.1109/EMBC.2018.8512732\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKansal K, Chandra TB, Singh A. ResNet-50 vs. EfficientNet-B0: Multi-Centric Classification of Various Lung Abnormalities Using Deep Learning. Procedia Comput Sci. 2024;235:70\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.procs.2024.04.007\u003c/span\u003e\u003cspan address=\"10.1016/j.procs.2024.04.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHwang JJ, Jung YH, Cho BH, Heo MS. An overview of deep learning in the field of dentistry. Imaging Sci Dent. 2019;49(1):1\u0026ndash;7. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5624/isd.2019.49.1.1\u003c/span\u003e\u003cspan address=\"10.5624/isd.2019.49.1.1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHasnain MA, Ali Z, Maqbool MS, Aziz M. X-ray Image Analysis for Dental Disease: A Deep Learning Approach Using EfficientNets. VFAST trans softw eng. 2024;12(3):147\u0026ndash;65. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.21015/vtse.v12i3.1912\u003c/span\u003e\u003cspan address=\"10.21015/vtse.v12i3.1912\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin S, Hao X, Liu Y, Yan D, Liu J, Zhong M. Lightweight deep learning methods for panoramic dental X-ray image segmentation. Neural Comput Applic. 2023;35(11):8295\u0026ndash;306. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00521-022-08102-7\u003c/span\u003e\u003cspan address=\"10.1007/s00521-022-08102-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAli MA, Fujita D, Kobashi S. Teeth and prostheses detection in dental panoramic X-rays using CNN-based object detector and a priori knowledge-based algorithm. Sci Rep. 2023;13(1):16542. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1038/s41598-023-43591-z\u003c/span\u003e\u003cspan address=\"10.1038/s41598-023-43591-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ealbumentations \u0026mdash; albumentations 1.1.0 documentation. Accessed November 10. 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://albumentations.readthedocs.io/en/latest/index.html\u003c/span\u003e\u003cspan address=\"https://albumentations.readthedocs.io/en/latest/index.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrossEntropyLoss. \u0026mdash; PyTorch 2.5 documentation. Accessed November 10, 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html\u003c/span\u003e\u003cspan address=\"https://pytorch.org/docs/stable/generated/torch.nn.CrossEntropyLoss.html\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYilmaz S, Tasyurek M, Amuk M, Celik M, Canger EM. Developing deep learning methods for classification of teeth in dental panoramic radiography. Oral Surg Oral Med Oral Pathol Oral Radiol. 2024;138(1):118\u0026ndash;27. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.oooo.2023.02.021\u003c/span\u003e\u003cspan address=\"10.1016/j.oooo.2023.02.021\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKurt A, G\u0026uuml;na\u0026ccedil;ar DN, Şılbır FY, et al. Evaluation of tooth development stages with deep learning-based artificial intelligence algorithm. BMC Oral Health. 2024;24(1):1034. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12903-024-04786-6\u003c/span\u003e\u003cspan address=\"10.1186/s12903-024-04786-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JH, Kim DH, Jeong SN. Diagnosis of cystic lesions using panoramic and cone beam computed tomographic images based on deep learning neural network. Oral Dis. 2020;26(1):152\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/odi.13223\u003c/span\u003e\u003cspan address=\"10.1111/odi.13223\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLeite AF, Gerven AV, Willems H, et al. Artificial intelligence-driven novel tool for tooth detection and segmentation on panoramic radiographs. Clin Oral Investig. 2021;25(4):2257\u0026ndash;67. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s00784-020-03544-6\u003c/span\u003e\u003cspan address=\"10.1007/s00784-020-03544-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMuramatsu C, Morishita T, Takahashi R, et al. Tooth detection and classification on panoramic radiographs for automatic dental chart filing: improved classification by multi-sized input data. Oral Radiol. 2021;37(1):13\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1007/s11282-019-00418-w\u003c/span\u003e\u003cspan address=\"10.1007/s11282-019-00418-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JH, Han SS, Kim YH, Lee C, Kim I. Application of a fully deep convolutional neural network to the automation of tooth segmentation on panoramic radiographs. Oral Surg Oral Med Oral Pathol Oral Radiol. 2020;129(6):635\u0026ndash;42. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.oooo.2019.11.007\u003c/span\u003e\u003cspan address=\"10.1016/j.oooo.2019.11.007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWeese J, Lorenz C. Four challenges in medical image analysis from an industrial perspective. Med Image Anal. 2016;33:44\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.media.2016.06.023\u003c/span\u003e\u003cspan address=\"10.1016/j.media.2016.06.023\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi P, Han T, Ren Y, Xu P, Yu H. Improved YOLOv4-tiny based on attention mechanism for skin detection. PeerJ Comput Sci. 2023;9:e1288. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.7717/peerj-cs.1288\u003c/span\u003e\u003cspan address=\"10.7717/peerj-cs.1288\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang J, Zhang J, Zhou K, Zhang Y, Chen H, Yan X. An Improved YOLOv5-Based Underwater Object-Detection Framework. Sens (Basel). 2023;23(7):3693. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/s23073693\u003c/span\u003e\u003cspan address=\"10.3390/s23073693\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-oral-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ohea","sideBox":"Learn more about [BMC Oral Health](http://bmcoralhealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/ohea/default.aspx","title":"BMC Oral Health","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Deep learning, Dental radiography, Intrabony lesions, Panoramic radiographs, Computer-aided diagnosis, Dental pathology, Machine learning, Neural networks, Image classification, Diagnostic imaging","lastPublishedDoi":"10.21203/rs.3.rs-6207574/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6207574/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eObjectives\u003c/h2\u003e \u003cp\u003eTo develop and evaluate an EfficientNet-B0-based deep-learning model for classifying eight types of intrabony lesions on dental panoramic radiographs.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA dataset of 833 dental panoramic radiographs from 245 patients was collected between October 2021 and April 2023 at two dental centers. Images were classified into eight categories: Periapical Widening, Condensing Osteitis, Periapical Granuloma, Nil Control, Diffuse Lesion, Periapical Abscess, Pericoronitis, and Radicular Cyst. Data preprocessing included class-weight computation and augmentation of minority classes. The EfficientNet-B0 model was trained for 50 epochs using the Adam optimizer with learning-rate scheduling and mixed-precision training.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eThe model achieved 93.04% validation accuracy, 0.9345 precision, 0.9304 recall, and 0.9295 F1-score. Performance analysis demonstrated robust classification across all lesion types, with the highest accuracy in Nil Control and Radicular Cyst identification.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe EfficientNet-B0 model demonstrates high accuracy in classifying dental intrabony lesions from panoramic radiographs, offering potential for enhanced diagnostic precision in clinical settings. Further validation across diverse clinical environments is recommended to establish a broader applicability.\u003c/p\u003e","manuscriptTitle":"EfficientNet-B0 Deep Learning Model for Accurate Classification of Intrabony Lesions in Dental Panoramic Radiographs","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-05 05:28:02","doi":"10.21203/rs.3.rs-6207574/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-07-29T08:53:17+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-07-20T11:01:13+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"168613162486106872862227014780450868887","date":"2025-07-03T18:28:59+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-05-12T12:30:15+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"315999658283170723876474997649564844965","date":"2025-05-07T10:15:43+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-04-02T07:36:34+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-04-01T08:35:08+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-04-01T06:31:48+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Oral Health","date":"2025-04-01T06:30:39+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-oral-health","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"ohea","sideBox":"Learn more about [BMC Oral Health](http://bmcoralhealth.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/ohea/default.aspx","title":"BMC Oral Health","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"c628a0a8-5a6f-4ca5-89e7-6eba978c2cce","owner":[],"postedDate":"May 5th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-10-23T08:53:20+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-05 05:28:02","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6207574","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6207574","identity":"rs-6207574","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}