A High-Accuracy Detection System: Based on Transfer Learning for Apical Lesions on Periapical Radiograph

Area of Science:

• Dentistry
• Medical Imaging
• Artificial Intelligence

Background:

• Apical lesions are common oral diseases often detected via periapical radiographs (PA).
• Manual marking of apical lesions in PAs is time-consuming for dentists during endodontic treatment.
• There is a need for automated tools to assist in the diagnosis of apical lesions.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN)-based model for automated detection of apical lesions in periapical radiographs.
• To improve the efficiency and accuracy of apical lesion diagnosis in endodontic practice.
• To reduce the manual workload for dentists in identifying apical lesions.

Main Methods:

• Development of a CNN-based regional analysis model for apical lesion detection.
• Implementation of an adaptive threshold preprocessing technique for image segmentation (accuracy >96%).
• Introduction of an enhanced technique for apical lesion symptom visualization.

Main Results:

• The proposed CNN model achieved a high detection accuracy of 96.21% for apical lesions.
• The model demonstrated a significant improvement of over 5% in accuracy compared to existing technologies.
• The adaptive threshold preprocessing technique achieved over 96% accuracy in image segmentation.

Conclusions:

• The developed CNN model effectively automates the diagnosis of apical lesions from periapical radiographs.
• This automation allows dentists to dedicate more time to critical aspects of patient care, such as treatment planning and communication.
• The study provides a valuable tool for enhancing diagnostic efficiency in endodontic procedures.