AI-based classification of odontogenic cysts using digital micrographs

1. Title:

Automated classification of four types of developmental odontogenic cysts

2. Authors:

Xiang Li; Yating Zhou; Jiawen Wang; et al.

3. Summary:

The researchers developed and evaluated a set of quantitative image features to automatically distinguish between four types of developmental odontogenic cysts (dentigerous, odontogenic keratocyst, lateral periodontal, and glandular) in H&E stained digital micrographs. Using support vector machine (SVM) and bagging with logistic regression (BLR) classifiers, the system achieved classification accuracy rates between 83.8% and 95.4%. The study serves as a proof-of-principle that quantitative features of cyst epithelia can be used for accurate diagnostic classification

4. Key Words:

Odontogenic cysts; Digital micrographs; Image features; Statistical classifiers; Artificial intelligence, Segmentation, Classification, Machine learning, Image processing

5. Extracted data

5.1. Year: 2014

5.2. Modality: Digital micrographs (histopathological images)

5.3. Dataset: 149 total images

5.4. Dataset Split: Training (73), Validation (37), and Testing (39)

5.5. Network Architecture: Support Vector Machine (SVM) and Bagging with Logistic Regression (BLR)

5.6. Metrics: TP rate (Recall), FP rate, Precision, F-measure, and ROC area

5.7. AP - Professional Qty: 1

5.8. AP - Supervisor Presence: Yes

5.9. AP - Experience Level: 31 years

5.10. AP - Expertise Area: Oral Maxillofacial Pathologist - OMFP

5.11. AP - Tool or System: No information

5.12. ML Task: Classification

5.13. Project Objective: To develop an automated epithelial classification algorithm that can distinguish between four types of odontogenic cysts based on histologic characteristics, ultimately reducing the workload of oral pathologists

6. Clinical Relevance: Proper diagnosis of odontogenic cysts is critical because different types exhibit different biological behaviors, require different treatment plans, and present varying levels of risk to patients. Because these pathologies are rare, diagnosis often requires subspecialists; therefore, an accurate computer-assisted diagnostic protocol could reduce the workload for oral pathologists and decrease the costs associated with seeking multiple expert opinions