Abstract
Computed tomography (CT) and X-ray images are indispensable diagnostic tools in modern dentistry for surgical planning and treatment. With the widespread adoption of dental cone beam CT (CBCT) in clinics, there is a significant opportunity to leverage three-dimensional (3D) data to reconstruct two-dimensional (2D) panoramic dental images. Such reconstructions provide a longitudinal view of the mandibular region without subjecting patients to additional X-ray exposure. This paper presents a fully automatic method for reconstructing 2D panoramic images using Bézier curves and Sequential Least Squares Programming (SLSQP) optimization. Tested on five patients, including those with missing teeth, the method successfully produced smooth, high-contrast panoramic images suitable for clinical diagnosis.
Introduction
In clinical dentistry, practitioners often require both 3D CT scans for precise surgical planning and 2D panoramic X-rays for a comprehensive, flattened view of the entire jaw. Currently, patients who have already undergone a 3D scan are frequently required to take a separate 2D panoramic X-ray. This practice increases the total radiation dose, contradicting the ALARA (As Low As Reasonably Achievable) principle of radiation safety.
While software-based "flattening" of 3D data into 2D images is possible, achieving high accuracy is challenging. The traditional "gold standard" involves a clinician manually tracing a curve over the 3D data to define the dental arch. However, manual definition is subjective, time-consuming, and often relies on simple geometric shapes like parabolas that fail to capture the complex, individual variations of a human jaw—particularly in patients with missing teeth or unique pathologies. This study addresses these limitations by introducing an automated, flexible optimization approach.
Methodology
The proposed computational pipeline automates the reconstruction process through four primary stages:
1. Segmentation and Pre-processing
The system imports DICOM files and applies global thresholding at 1,500 Hounsfield Units to isolate high-density dental enamel. It automatically identifies the axial slice containing the highest density of segmented voxels, which typically corresponds to the tips of the teeth.
2. Skeletonization
To ensure the algorithm is robust against gaps caused by missing teeth, the method employs morphological dilation and Gaussian smoothing. This creates a continuous "mask" of the jaw. A medial axis algorithm is then applied to extract the "skeleton," or the center line of this mask, providing the target points for curve fitting.
3. Bézier Optimization
The core innovation lies in fitting a Bézier curve to the extracted skeleton. Unlike rigid parabolas, Bézier curves offer the flexibility needed to match individual anatomy. The researchers utilized Sequential Least Squares Programming (SLSQP) to minimize the distance between the Bézier curve and the skeleton points. The performance was evaluated using 3, 7, 11, and 15 control points to determine the optimal balance between simplicity and accuracy.
4. Image Synthesis
Once the optimal curve is established, the system generates parallel curves to account for the thickness of the dental arch. Average Intensity Projection (AIP) is then used to extract and flatten the pixels along these curves into the final 2D panoramic image.
Results and Discussion
The study utilized CBCT datasets from five different patients, sourced from various hardware to ensure the method's independence from specific CT machines. Notably, the algorithm proved robust in cases involving missing teeth.
Quantitative and Qualitative Performance
Quantitative analysis showed that the "cost function"—the average distance between the optimized curve and the actual dental skeleton—dropped drastically as the number of control points increased. For instance, in one patient, the cost decreased from 5.70 (using 3 points) to 0.87 (using 15 points).
Qualitatively, images generated with 11 and 15 control points were significantly clearer and more detailed. These higher-order curves captured nuances such as tooth root structures that were blurred or entirely missing when using fewer control points (the "parabola problem"). The flexibility of the Bézier function allowed the reconstruction to follow the mandibular region almost perfectly.
Conclusion
This work demonstrates that Bézier curve optimization provides a computationally efficient and highly accurate method for dental image reconstruction. By automating the generation of 2D panoramic views from existing 3D CBCT data, the method offers a "free" diagnostic tool that improves clinical workflow and adheres to radiation safety standards. Future work will focus on validating the method against a broader range of pathologies, such as impacted teeth and significant jaw deformities, and exploring alternative parametric curves like B-splines for further refinement.
Summary updated