Optical tomographic imaging of activation of the infant auditory cortex using perturbation Monte Carlo with anatomical a priori information

Juha Heiskala, Kalle Kotilahti, Lauri Lipiäinen, Petri Hiltunen, P. Ellen Grant, Ilkka NissiläProceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE 2007 · DOI

TL;DR

This paper introduces a novel perturbation Monte Carlo (pMC) method for optical tomography that leverages anatomical prior information to improve the accuracy of imaging brain activation in infants. By incorporating a personalized, albeit derived from another infant, anatomical model, the pMC approach significantly enhances reconstruction quality compared to homogeneous models, paving the way for more precise non-invasive brain monitoring in this vulnerable population.

What Problem Does This Paper Address?

Optical tomography is a promising non-invasive technique for monitoring brain activity by measuring changes in light absorption and scattering. However, accurately reconstructing these changes, especially in the complex and dynamic infant brain, is challenging. A key hurdle is the need for precise anatomical information to guide the reconstruction process. Standard methods often rely on simplified homogeneous models, which can lead to inaccurate localization and quantification of brain activity. This paper tackles the problem of improving the accuracy of optical tomography reconstructions by integrating anatomical prior information, specifically for imaging hemodynamic changes related to brain activation in infants. The challenge lies in obtaining this anatomical information for individual infants and effectively incorporating it into the imaging model.

Methods

The core of this work is the development and application of a perturbation Monte Carlo (pMC) method for frequency-domain optical tomography. Here's a breakdown of the key techniques:

  • Monte Carlo (MC) Simulation: This forms the basis of the forward model, simulating the complex path of photons through tissue. The pMC approach specifically focuses on calculating the change in signal due to a perturbation, making it computationally more feasible for inverse problems than full MC simulations for every possible perturbation.
  • Frequency-Domain Data: The method utilizes measurements of both the amplitude and phase shift of light signals, which provide complementary information for reconstruction.
  • Anatomical Prior Information: This is a crucial element. The authors use T1-weighted MR images from neonates. Since direct MR imaging of the study infants was not feasible, they employ a deformation technique to warp a pre-existing anatomical model (from another infant) to match the surface geometry of the infant being studied. This is achieved by digitizing landmark points on the infant's scalp and using these to guide the deformation.
  • Voxel-Based Model: The anatomical model is represented as a 3D grid of voxels, allowing for the assignment of different optical properties to various tissue types (scalp, skull, gray matter, white matter, CSF).
  • Inverse Problem Solution: The reconstruction of absorption changes is achieved by minimizing the squared difference between simulated and experimental data (both intensity and phase shift). This is done iteratively using a gradient-based approach.
  • Comparison with FEM: For validation and comparison, the authors also employ a finite element method (FEM) based approach using the TOAST software, which also utilizes anatomical models.

What's novel is the specific application of pMC with a deformed anatomical model derived from another infant, demonstrating the utility of generic infant brain atlases.

Datasets

  • Simulated Data: The primary "dataset" used for evaluation was generated through simulations. A digital phantom (an anatomical model with known optical properties) was created, and simulated measurements were performed under conditions mimicking brain activation. This allowed for a controlled assessment of the reconstruction accuracy.
  • Experimental Data: The method was also tested on experimental data from three healthy neonates. This data consisted of frequency-domain optical measurements taken while the infants were exposed to auditory stimuli.

The paper does not introduce a new benchmark dataset but rather uses simulated data for rigorous evaluation and experimental data for real-world validation.

Key Insights and Ideas

  • Anatomical Priors are Crucial: The most significant finding is that incorporating anatomical information, even if derived from a different infant and warped to match the subject, dramatically improves the accuracy and localization of reconstructed absorption changes. This highlights the power of "a priori" knowledge in optical tomography.
  • Phase Information Matters: Using both intensity and phase shift data in the pMC reconstruction leads to a more localized and higher-contrast representation of the perturbation compared to using intensity data alone. This is particularly important when the perturbation is subtle or falls between measurement points.
  • Generic Atlases are Viable: The success of using a deformed anatomical model from another infant suggests that generic infant brain atlases can be valuable for optical imaging studies, reducing the reliance on individual MR scans for every subject.
  • pMC vs. FEM: While both methods benefit from anatomical information, the pMC method, when using a heterogeneous anatomical model, produced more localized reconstructions compared to the FEM approach with a homogeneous background.

Key Results

  • Improved Localization: Reconstructions using the pMC method with a deformed heterogeneous anatomical model showed significantly better localization of simulated absorption perturbations compared to homogeneous models.
  • Enhanced Contrast: The contrast of the reconstructed perturbation was higher and closer to the true value when both phase and intensity data were used in the pMC method.
  • Robustness to Errors: The pMC method with the deformed anatomical model remained reasonably accurate even when the optical properties were off by 10% from the true values.
  • Experimental Validation: The experimental data showed measurable changes in signal intensity and phase shift in response to auditory stimuli, supporting the feasibility of the approach for studying infant brain activity.
  • FEM vs. pMC (Experimental): In experimental data, the pMC method produced a more localized reconstruction of oxygenated hemoglobin changes, with the bulk of the change in superficial regions, whereas the FEM solution was more spread out.

Limitations and Future Work

  • Anatomical Model Deformation: While effective, the deformation of the anatomical model is a rudimentary approach. Future work could involve more refined deformation algorithms that better preserve essential anatomical properties.
  • Averaging Anatomical Information: The authors suggest that developing methods for averaging anatomical information from multiple subjects could further improve results.
  • Noise Sensitivity: The paper notes that in lower contrast-to-noise situations, noise might reduce the effectiveness of phase information. Further investigation into noise reduction and its impact on phase-based reconstruction is warranted.
  • Limited Experimental Data: The experimental validation was performed on a small group of three infants. Expanding the experimental dataset would strengthen the findings.
  • Focus on Absorption: The current work primarily focuses on absorption changes. Future research could explore incorporating scattering changes as well.

Summary updated

Abstract

We have developed a perturbation Monte Carlo method for calculating forward and inverse solutions to the optical tomography imaging problem in the presence of anatomical <i>a priori</i> information. The method uses frequency domain data. In the present work, we consider the problem of imaging hemodynamic changes due to brain activation in the infant brain. We test finite element method and Monte Carlo based implementations using a homogeneous model with the exterior of the domain warped to match digitized points on the skin. With the perturbation Monte Carlo model, we also test a heterogeneous model based on anatomical <i>a priori</i> information derived from a previously recorded infant T1 magnetic resonance (MR) image. Our simulations show that the anatomical information improves the accuracy of reconstructions quite significantly even if the anatomical MR images are based on another infant. This suggests that significant benefits can be obtained by the use of generic infant brain atlas information in near-infrared spectroscopy and optical tomography studies.
BibTeX
@inproceedings{heiskala2007optical,
  author = {Juha Heiskala and Kalle Kotilahti and Lauri Lipiäinen and Petri Hiltunen and P. Ellen Grant and Ilkka Nissilä},
  title = {{Optical tomographic imaging of activation of the infant auditory cortex using perturbation Monte Carlo with anatomical a priori information}},
  booktitle = {Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE},
  year = {2007},
  doi = {10.1117/12.728276},
}
Figure