Improving diffuse optical tomography imaging quality using APU-Net: an attention-based physical U-Net model.

Area of Science:

• Medical Imaging
• Biomedical Optics
• Artificial Intelligence in Medicine

Background:

• Traditional diffuse optical tomography (DOT) reconstructions suffer from image artifacts due to factors like source proximity, poor optode coupling, tissue heterogeneity, and shadowing effects from large lesions.
• These artifacts compromise DOT image quality, hindering accurate lesion diagnosis and impacting clinical utility.

Purpose of the Study:

• To introduce an attention-based U-Net (APU-Net) model for enhancing DOT image reconstruction quality.
• To address specific DOT reconstruction challenges, including artifact-induced distortions and lesion-shadowing effects.
• To improve lesion diagnostic accuracy through enhanced DOT imaging.

Main Methods:

• Developed an APU-Net model integrating a contextual transformer attention module for DOT reconstruction.
• Trained the APU-Net model using both simulation and phantom data, specifically targeting artifact reduction and shadowing effect mitigation.
• Validated the model's performance on clinical patient data.

Main Results:

• The APU-Net model demonstrated significant artifact reduction, with an average artifact contrast decrease of 26.83%.
• The model improved overall image quality and enhanced contrast depth profiles, showing average increases of 20.28% and 45.31% for the second and third target layers, respectively.
• Clinical data evaluation confirmed the model's efficacy in improving breast cancer diagnosis.

Conclusions:

• The APU-Net model effectively enhances DOT image quality by minimizing reconstruction artifacts.
• The improved target depth profile achieved by APU-Net aids in more accurate lesion characterization.
• This AI-driven approach shows promise for advancing DOT applications in clinical diagnostics, particularly for breast cancer detection.