Automatic tooth roots segmentation of cone beam computed tomography image sequences using U-net and RNN

Area of Science:

• Dental Imaging and Diagnostics
• Medical Image Analysis
• Artificial Intelligence in Healthcare

Background:

• Automatic tooth root segmentation is crucial for 3D dental model reconstruction from Cone Beam Computed Tomography (CBCT) images.
• Manual segmentation is time-consuming and challenging due to similar gray values between tooth roots and alveolar bone in CBCT scans.

Purpose of the Study:

• To develop and evaluate an automatic tooth root segmentation algorithm for CBCT axial image sequences using deep learning.
• To address the limitations of manual segmentation in dental imaging.

Main Methods:

• A novel deep learning method combining U-net with attention gates (AGs) and a Recurrent Neural Network (RNN) was proposed.
• The RNN was utilized to capture contextual information from adjacent CBCT slices, enhancing segmentation accuracy.
• The model was trained and tested on a substantial dataset of 1591 CBCT images.

Main Results:

• The proposed method achieved high segmentation accuracy on the testing dataset.
• Key performance metrics included Intersection over Union (IOU) of 0.914, Dice Similarity Coefficient (DICE) of 0.955, Average Precision Rate (APR) of 95.8%, Average Recall Rate (ARR) of 95.3%, and Average Symmetrical Surface Distance (ASSD) of 0.145 mm.

Conclusions:

• The integration of attention U-net and RNN demonstrates promising results for automatic tooth root segmentation.
• This deep learning approach has the potential to enhance efficiency and accuracy in clinical dental diagnosis and treatment planning.