Deep Learning-Based Multiclass Instance Segmentation for Dental Lesion Detection.

Area of Science:

• Artificial Intelligence in Dentistry
• Medical Image Analysis
• Computer Vision

Background:

• Manual dental disease assessment from X-rays is time-consuming and prone to errors, especially for less experienced dentists.
• Advanced computer vision, machine learning, and deep learning models are increasingly used for automated dental disease diagnosis.
• Periapical lesions require accurate and efficient detection for timely treatment.

Purpose of the Study:

• To propose a lightweight Mask-RCNN model for accurate periapical disease detection and localization in dental X-ray images.
• To develop an efficient AI model that can assist dental clinicians in diagnosing periapical lesions.
• To evaluate the model's performance on a custom annotated dataset of various periapical lesions.

Main Methods:

• A lightweight Mask-RCNN model was developed, featuring a modified MobileNet-v2 backbone and a region-based network (RPN).
• The model was designed for efficient periapical disease localization, particularly effective on smaller datasets.
• Performance was evaluated on a custom dataset containing five types of periapical lesions.

Main Results:

• The proposed lightweight Mask-RCNN model achieved an overall accuracy of 94% for periapical lesion detection.
• The model demonstrated a mean average precision (mAP) of 85% and a mean intersection over union (mIoU) of 71.0%.
• Significant improvements in detection, classification, and localization accuracy were observed compared to existing methods, using fewer images.

Conclusions:

• The lightweight Mask-RCNN model offers a highly accurate and efficient solution for automated periapical disease detection in dental radiography.
• This AI approach enhances diagnostic capabilities, potentially reducing errors and improving patient outcomes.
• The model outperforms state-of-the-art methods, demonstrating its potential for clinical application with smaller datasets.