AI Differentiates Ameloblastoma vs OKC on Multidetector CT
1. Title Computer tomographic differential diagnosis of ameloblastoma and odontogenic keratocyst: classification using a convolutional neural network
2. Authors Mayara Simões Bispo; Mário Lúcio Gomes de Queiroz Pierre Júnior; Antônio Lopes Apolinário Jr; Jean Nunes dos Santos; Braulio Carneiro Junior; Frederico Sampaio Neves; Iêda Crusoé-Rebello.
3. Summary
This study evaluated the automatic classification performance of a Convolutional Neural Network (CNN), specifically Google Inception v3, using multidetector computed tomography (MDCT) images to differentiate odontogenic keratocysts (OKCs) and ameloblastomas (AMs). A total of 350 original axial CT images (22 AM and 18 OKC cases) were manually segmented and augmented to 2,500 images. A 2-fold × 5 cross-validation method was applied. The CNN achieved classification accuracies above 90% in all five iterations, with a higher misclassification rate for ameloblastomas.
VISUAL SUMMARY
4. Key Words Ameloblastoma; Odontogenic Cysts; Artificial Intelligence; Tomography; X-Ray Computed.
5. Extracted Data
5.1 Year 2021
5.2 Modality Multidetector Computed Tomography (MDCT) – Axial CT images
5.3 Dataset 350 axial images
5.4 Dataset Split 70% Development set / 30% Final test set / 2-fold × 5 cross-validation
5.5 Network Architecture Inception v3 (CNN)
5.6 Metrics Accuracy; Precision; Confusion Matrix ; Sensibility; F1-Score
5.7 AP - Professional Qty 1
5.8 AP - Supervisor Presence No information
5.9 AP - Experience Level Experienced examiner (years not specified)
5.10 AP - Expertise Area No information
5.11 AP - Tool or System ImageJ
5.12 ML Task Binary classification (AM vs OKC)
5.13 Project Objective To evaluate CNN performance in differentiating AM and OKC using MDCT images.
6. Clinical Relevance Accurate differentiation between ameloblastoma and odontogenic keratocyst is essential due to different treatment approaches. The CNN achieved >90% accuracy, suggesting potential as an auxiliary diagnostic tool in oral radiology.
