Quantitative evaluation of atlas-based high-density diffuse optical tomography for imaging of the human visual cortex

Area of Science:

• Biomedical Optics
• Medical Imaging
• Neuroscience

Background:

• Accurate light propagation models are crucial for diffuse optical tomography (DOT) brain imaging.
• Subject-specific models improve image reconstruction, but atlas-based models offer a promising alternative when unavailable.
• Limited analysis exists on the relationship between geometrical accuracy, light propagation errors, and dynamic imaging in DOT.

Purpose of the Study:

• To evaluate the impact of geometrical errors from rigid registration algorithms on DOT image reconstruction.
• To assess the performance of 11 different rigid registration algorithms in atlas-based DOT studies.
• To determine the feasibility of using atlas-based models for dynamic brain imaging in DOT.

Main Methods:

• Simulated 24 subjects with varying anatomical features.
• Applied 11 distinct rigid registration algorithms to create atlas-based models.
• Analyzed the correlation between geometrical surface errors and internal light propagation errors.
• Quantified the localization error of recovered focal activations in the visual cortex.

Main Results:

• A strong correlation (R² = 0.97) was observed between geometrical surface error and internal light propagation errors.
• Despite correlations, the overall variation in recovered focal activations was minimal.
• Subject-specific models yielded the best results with an average location error of 1.2 mm.
• No evaluated rigid algorithm resulted in localization errors exceeding 4.5 mm.

Conclusions:

• Rigid registration algorithms are a viable and promising approach for atlas-based DOT imaging when subject-specific models are not feasible.
• Atlas-based DOT can achieve acceptable accuracy for dynamic brain imaging, reducing reliance on individualized models.
• The study provides valuable insights into the robustness of DOT image reconstruction against geometrical inaccuracies in atlas-based approaches.