Difference imaging from single measurements in diffuse optical tomography: a deep learning approach.

Area of Science:

• Medical Imaging
• Biomedical Optics
• Artificial Intelligence in Medicine

Background:

• Difference imaging in diffuse optical tomography (DOT) is crucial for in vivo imaging.
• Traditional DOT difference imaging requires additional reference measurements, which is time-consuming.
• Mismatches between target and reference measurements can lead to inaccurate DOT reconstructions.

Purpose of the Study:

• To streamline DOT data acquisition by eliminating reference measurements.
• To mitigate reconstruction inaccuracies caused by target-reference medium mismatches.
• To improve the efficiency and accuracy of DOT difference imaging using deep learning.

Main Methods:

• Developed a deep learning model (artificial neural network) to generate difference imaging data from target measurements only.
• Trained and validated the model using simulation data.
• Tested the model on simulations, phantom experiments, and clinical data from 56 breast lesion patients.

Main Results:

• The deep learning method achieved performance comparable to traditional methods without mismatch.
• It outperformed traditional methods when target-reference mismatches were present.
• Demonstrated improvements in target-to-artifact ratio and lesion localization in patient data.

Conclusions:

• The proposed deep learning approach simplifies DOT data acquisition.
• It effectively mitigates problems associated with measurement mismatches.
• It enhances reconstructed image quality for DOT difference imaging.