Semi-supervised Teeth Segmentation

Computer-aided diagnosis tools are increasingly popular in modern dental practice, particularly for treatment planning or comprehensive prognosis evaluation. In particular, the 2D panoramic X-ray image is an efficient way for dentists to determine invisible caries, impacted teeth and supernumerary teeth among children. Additionally, the 3D dental cone-beam computed tomography (CBCT) examinations are extensively utilized in orthodontics and endodontics due to their low ray dose and the capability to visualize three-dimensional structures. However, identifying teeth from panoramic X-ray images or CBCT scans and further manually annotating the teeth is time-consuming and labor-intensive. Thus, we usually cannot obtain a huge number of labeled cases, which limits the development of deep learning algorithms for segmenting teeth and automatically analyzing diseases. As a potential alternative, semi-supervised learning can explore useful information from unlabeled cases.

In this challenge, we aim to promote the development of the tooth segmentation in panoramic X-ray images. We extend the Semi-TeethSeg2023 Challenge from binary semantic settings to multi-instance and multi-class segmentation settings, with an emphasis on instance-level segmentation. Specifically, in Semi-TeethSeg2024, we will provide instance annotations for different teeth, inclusive of pertinent category information. The segmentation algorithm is expected to accurately segment 32 permanent teeth (including wisdom teeth) and 20 possible deciduous teeth in panoramic X-ray images.

In contrast to the dataset in Semi-TeethSeg2023 (6500 panoramic X-ray images and 584 CBCT scans), the dataset for Semi-TeethSeg2024 has been expanded to include 7000 panoramic X-ray images and 600 CBCT scans. To the best knowledge, this will be the largest most diverse publicly available dataset for teeth instance segmentation.  In addition, based on the results in Semi-TeethSeg2023, we found that segmentation models can not achieve a good tradeoff between segmentation accuracy and efficiency. Thus, in Semi-TeethSeg2024, the challenge evaluation criteria are not limited to segmentation accuracy, but also include runtime and GPU memory consumption,
providing a comprehensive assessment of segmentation efficiency.

In summary, the Semi-TeethSeg2024 challenge has three main features:
(1) Task: this is the first challenge for teeth instance segmentation in panoramic X-ray images and CBCT scans.
(2) Dataset: we provide the largest dataset, including 7000 2D X-ray images and 600 3D CBCT scans spanning all age groups.
(3) Evaluation: we not only focus on segmentation accuracy but also segmentation efficiency, which are in concordance with real clinical practice and requirements.