Towards real-time diffuse optical tomography with a handheld scanning probe DOI Creative Commons
Robin Dale, Nicholas Ross, Scott S. Howard

et al.

Biomedical Optics Express, Journal Year: 2025, Volume and Issue: 16(4), P. 1582 - 1582

Published: March 14, 2025

Diffuse optical tomography (DOT) performed using deep-learning allows high-speed reconstruction of tissue properties and could thereby enable image-guided scanning, e.g., to enhance clinical breast imaging. Previously published models are geometry-specific and, therefore, require extensive data generation training for each use case, restricting the scanning protocol at point use. A transformer-based architecture is proposed overcome these obstacles that encode spatially unstructured DOT measurements, enabling a single trained model handle arbitrary pathways measurement density. The demonstrated with tissue-emulating simulated phantom data, yielding - 24 mm-deep absorptions (μ ) reduced scattering s ') images, respectively average RMSEs 0.0095±0.0023 cm-1 1.95±0.78 cm-1, Sørensen-Dice coefficients 0.55±0.12 0.67±0.1, anomaly contrast 79±10% 93.3±4.6% ground-truth contrast, an effective imaging speed 14 Hz. absolute μ ' values homogeneous examples were within 10% true values.

Language: Английский

Towards real-time diffuse optical tomography with a handheld scanning probe DOI Creative Commons
Robin Dale, Nicholas Ross, Scott S. Howard

et al.

Biomedical Optics Express, Journal Year: 2025, Volume and Issue: 16(4), P. 1582 - 1582

Published: March 14, 2025

Diffuse optical tomography (DOT) performed using deep-learning allows high-speed reconstruction of tissue properties and could thereby enable image-guided scanning, e.g., to enhance clinical breast imaging. Previously published models are geometry-specific and, therefore, require extensive data generation training for each use case, restricting the scanning protocol at point use. A transformer-based architecture is proposed overcome these obstacles that encode spatially unstructured DOT measurements, enabling a single trained model handle arbitrary pathways measurement density. The demonstrated with tissue-emulating simulated phantom data, yielding - 24 mm-deep absorptions (μ ) reduced scattering s ') images, respectively average RMSEs 0.0095±0.0023 cm-1 1.95±0.78 cm-1, Sørensen-Dice coefficients 0.55±0.12 0.67±0.1, anomaly contrast 79±10% 93.3±4.6% ground-truth contrast, an effective imaging speed 14 Hz. absolute μ ' values homogeneous examples were within 10% true values.

Language: Английский

Citations

0