A Stage-Wise Residual Attention Generation Adversarial Network for Mandibular Defect Repairing and Reconstruction

Area of Science:

• Oral and Maxillofacial Surgery
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Mandibular defects pose significant challenges in oral-maxillofacial surgery, impacting facial aesthetics and function.
• Current digitalized design methods for reconstruction have limitations including manual operations and high error rates.
• There is a critical need for automated, precise, and individualized reconstruction techniques.

Purpose of the Study:

• To develop and evaluate an automated method for precise mandibular defect reconstruction.
• To address the limitations of traditional digitalized design methods in oral-maxillofacial surgery.
• To improve the accuracy and applicability of mandibular reconstruction techniques.

Main Methods:

• Proposed a Stage-wise Residual Attention Generative Adversarial Network (SRA-GAN) for mandibular defect reconstruction.
• Implemented a stage-wise residual attention mechanism in the generator to improve spatial information extraction.
• Utilized a multi-field perceptual network with parallel discriminators and a self-encoder perceptual loss function to minimize errors and ensure anatomical correctness.

Main Results:

• The SRA-GAN achieved a Dice Similarity Coefficient (DSC) of 94.238% on a custom mandibular defect dataset.
• The method demonstrated a 95% Hausdorff Distance (HD95) of 4.787, indicating high accuracy.
• Experimental results show promising prospects for clinical application in mandibular reconstruction.

Conclusions:

• The proposed SRA-GAN offers an automated, precise, and individualized solution for mandibular defect reconstruction.
• This AI-driven approach overcomes the limitations of existing methods, reducing reconstruction errors.
• The SRA-GAN shows significant potential for enhancing patient outcomes in oral-maxillofacial surgery.