Related Experiment Video
Updated: Jun 28, 2026
09:16
Supervised Machine Learning for Semi-Quantification of Extracellular DNA in Glomerulonephritis
Published on: June 18, 2020
6.9K
Classification of Ameloblastoma, Periapical Cyst, and Chronic Suppurative Osteomyelitis with Semi-Supervised
Bohui Liang1, Hongna Qin2, Xiaolin Nong3
1School of Computer, Electronics and Information, Guangxi University, Nanning 530004, China.
Bioengineering (Basel, Switzerland)
|June 27, 2024
Summary
A new WaveletFusion-ViT model accurately diagnoses maxillofacial diseases like ameloblastoma from CBCT scans using semi-supervised learning. This approach achieves high accuracy with limited labeled data, outperforming fully supervised methods.
Area of Science:
- Dentistry
- Medical Imaging
- Artificial Intelligence
Background:
- Ameloblastoma, periapical cysts, and chronic suppurative osteomyelitis present similar imaging features on CBCT scans, complicating differential diagnosis.
- Current deep learning diagnostic models often necessitate manual region delineation, posing practical challenges.
Purpose of the Study:
- To develop an automated diagnostic model for distinguishing between ameloblastoma, periapical cysts, and chronic suppurative osteomyelitis using CBCT panoramic images.
- To address the limitations of manual region-of-interest annotation in existing deep learning approaches.
Main Methods:
- A novel Wavelet Extraction and Fusion Module with Vision Transformer (WaveletFusion-ViT) model was proposed.
- The model utilized 539 CBCT samples (healthy, AM, PC, CSO) and 2000 healthy samples for pre-training a domain-adaptive network (DAN).
- Semi-supervised learning (SSL) methods were employed to train the WaveletFusion-ViT model, initialized with pre-trained DAN weights.
Main Results:
- The WaveletFusion-ViT model achieved an average accuracy of 91.47% and an AUC of 0.942 across five-fold cross-validation.
- The model demonstrated 91.47% accuracy using less than 20% of labeled data, surpassing the fully supervised approach's 89.05% accuracy.
- High sensitivity (79.60%) and specificity (94.48%) were recorded.
Conclusions:
- The WaveletFusion-ViT model offers an effective and automated method for diagnosing prevalent maxillofacial diseases from CBCT images.
- The study highlights the potential of combining SSL with deep learning for accurate medical image diagnosis using minimal labeled data.
- Future research should address the limited number of CSO cases and improve diagnostic accuracy for this condition.
