International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 04 | Apr 2025
p-ISSN: 2395-0072
www.irjet.net
Advanced U-Net-Based Semantic Segmentation for Panoramic Dental X-Ray Analysis Ms. Femil Rence L1, Mrs. Grace Berin T2 & Mrs. Anusha K3 12nd year PG student, Ponjesly College of Engineering 2Assistant Professor, Ponjesly College of Engineering
3Assistant Professor, Ponjesly College of Engineering
***
Abstract — Medical image analysis in dental diagnostics encounters significant challenges, particularly in precisely segmenting anatomical components within panoramic X-ray images. This research investigates the use of U-Net-based deep learning architectures for semantic segmentation of dental structures. Six U-Net variations, including Vanilla U-Net, Dense U-Net and Attention U-Net, were examined to evaluate their segmentation performance. The dataset comprised panoramic dental X-ray images, which underwent pre-processing through Contrast Limited Adaptive Histogram Equalization (CLAHE) and resizing to ensure efficient training. Performance metrics such as Dice Coefficient, Intersection over Union (IoU), F1 Score and Accuracy were utilized for model evaluation. Results demonstrated that Vanilla U-Net with two convolutional layers achieved an effective balance between accuracy and computational cost. Future research aims to further enhance segmentation precision through architectural advancements and dataset enrichment.
Key
Words: Semantic Segmentation, Dice Coefficient, IoU, F1 Score.
U-Net,
Panoramic
Dental
X-Ray,
Deep
Learning,
Image
Processing,
1. INTRODUCTION Oral health plays a vital role in maintaining overall wellness, and the early detection of dental issues like cavities, gum diseases, and bone loss is essential for successful treatment. Panoramic dental X-ray imaging offers a detailed overview of oral structures, making it a valuable diagnostic resource. However, conventional analysis of these images primarily depends on manual evaluation by dentists and radiologists, which can be time-consuming and susceptible to human error. Additionally, factors like varying image quality, differences in patient anatomy, and overlapping features make precise diagnosis more challenging. In recent times, Artificial Intelligence (AI) and deep learning methods have demonstrated remarkable potential in automating medical image analysis and improving its accuracy [1], [2]. These AI-based systems have already shown success in cancer identification [3], lung disease prediction [4], and personalized treatments [5], highlighting their capability to transform healthcare. In the case of medical imaging, semantic segmentation models have gained substantial attention among deep learning approaches. U-Net stands out as a widely adopted architecture which is focused and designed for biomedical image segmentation tasks. It features two major things a encoder and decoder framework combined with skip connections, enabling accurate pixel-level classification while retaining spatial information [6]. Over time, various enhanced versions of U-Net, including Dense, Attention, Residual, and Nested, have been introduced to boost segmentation accuracy, improve feature transmission, and increase computational effectiveness [7]. Additionally, lightweight deep learning models have been explored to enhance efficiency while maintaining high accuracy in panoramic dental X-ray segmentation [8]. AI-based models have also been utilized in various medical imaging fields, including transformer-based DL networks for dental segmentation [9] and CNN-based automated analysis of dental implants in CBCT images [10]. This study focuses on evaluating and comparing six U-Net variants for segmenting panoramic dental X-ray images to identify the most effective model for clinical applications. The dataset undergoes pre-processing, annotation, model training, and validation, with segmentation performance evaluated using Dice Coefficient, Intersection over Union (IoU), F1 Score and Accuracy. Previous research has demonstrated the effectiveness of multi-scale feedback feature refinement U-Net for medical image segmentation [11] and the application of attention mechanisms in U-Net for enhanced segmentation [12]. Additionally, advancements in computer-aided diagnostic (CAD) systems have facilitated the integration of AI-driven segmentation with real-time clinical decision-making tools [13]. By identifying the optimal U-Net variant for dental X-ray segmentation, this
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 1048