Related Experiment Video
Updated: May 2, 2026
Deep Learning-Based Segmentation of Cryo-Electron Tomograms
Published on: November 11, 2022
Automatic classification and segmentation of multiclass jaw lesions in cone-beam CT using deep learning
Wei Liu1, Xiang Li2, Chang Liu1
1State Key Laboratory of Oral Diseases & National Center for Stomatology & National Clinical Research Center for Oral Diseases & Department of Oral and Maxillofacial Surgery, West China Hospital of Stomatology, Sichuan University, Chengdu 610041, China.
A deep learning model accurately classifies and segments jaw lesions on cone-beam CT scans, outperforming human experts and improving their diagnostic accuracy. This AI tool enhances efficiency in distinguishing complex lesions like ameloblastoma and odontogenic keratocysts.
Area of Science:
- Oral and Maxillofacial Radiology
- Artificial Intelligence in Medical Imaging
- Deep Learning for Medical Diagnosis
Background:
- Accurate classification and segmentation of jaw lesions using cone-beam CT (CBCT) are crucial for effective treatment planning.
- Manual interpretation of CBCT scans can be time-consuming and subject to inter-observer variability.
- Deep learning (DL) models offer potential for automating and improving the accuracy of medical image analysis.
Purpose of the Study:
- To develop and validate a modified deep learning (DL) model based on nnU-Net for classifying and segmenting five types of jaw lesions.
- To evaluate the model's performance against oral and maxillofacial surgeons (OMSs) and radiologists (OMFRs).
- To assess the impact of AI assistance on the diagnostic performance and efficiency of clinicians.
Main Methods:
- A multi-class segmentation DL model was trained using 368 CBCT scans (37,168 slices) with manual annotations by OMSs as ground truth.
- Performance metrics included sensitivity, specificity, precision, F1-score, and accuracy for classification, and Dice Similarity Coefficient (DSC), Average Symmetric Surface Distance (ASSD), and segmentation time for segmentation.
- The model's classification and segmentation capabilities were compared to those of OMSs and OMFRs, with and without AI assistance.
Main Results:
- The DL model achieved high classification performance (sensitivity 0.871, specificity 0.974, precision 0.874, accuracy 0.891), surpassing human experts.
- AI assistance significantly improved classification performance for OMSs and OMFRs, especially for differentiating challenging lesions like ameloblastoma (AM) and odontogenic keratocyst (OKC) (F1-score improvement: 6.2%–12.7%).
- The model demonstrated strong segmentation performance (DSC 87.2%, ASSD 1.359 mm) and drastically reduced segmentation time (40 ± 9.9 s) compared to manual segmentation (25 ± 7.2 min).
Conclusions:
- The developed DL model provides accurate and efficient classification and segmentation of five jaw lesion types in CBCT images.
- The AI tool effectively assists clinicians in improving diagnostic accuracy and segmentation efficiency, particularly for difficult-to-distinguish lesions.
- This technology holds significant promise for enhancing diagnostic workflows in oral and maxillofacial radiology.
