Abstract
When performing diffuse optical tomography (DOT) of the breast, mismatches between forward models and experimental conditions often degrade reconstruction accuracy. Traditionally, reference measurements using physical phantoms are required for calibration, but customizing these phantoms for individual patients is clinically impractical. This paper proposes a data self-calibration method leveraging a high-density (HD) parallel-plate DOT system. By exploiting geometric symmetry and redundant measurements at identical source-detector (SD) distances, a virtual reference can be constructed directly from patient data. Validated through Monte Carlo simulations, phantom experiments, and clinical trials, this approach maintains reconstruction quality while significantly simplifying the clinical workflow.
The Challenge of Calibration in Breast DOT
Diffuse Optical Tomography (DOT) is a non-invasive imaging modality that uses near-infrared light to map functional tissue properties, such as hemoglobin concentration. A significant hurdle in clinical DOT is the discrepancy between theoretical "forward models" and actual experimental data, caused by factors like sensor coupling variability and unique patient anatomy.
To mitigate this, clinicians typically use "relative measurements," where data from the patient is compared against data from a physical reference phantom that mimics the patient's breast shape and optical background. However, creating custom phantoms for every patient is slow and expensive, hindering the widespread adoption of DOT in high-throughput clinical settings.
Methods: Exploiting Geometric Redundancy
The proposed Data Self-Calibration Method removes the need for physical phantoms by utilizing the high sensor density of a parallel-plate DOT system.
The core innovation lies in three areas:
- Geometric Symmetry: The system employs a 7x8 array of LEDs and photodiodes. Due to the parallel-plate arrangement, numerous SD pairs share the exact same Euclidean distance, providing significant data redundancy.
- Virtual Reference Construction: Assuming that tumors are small relative to the breast volume, the researchers hypothesized that for any given SD distance, some light paths will traverse only healthy, homogeneous tissue. By grouping measurements by distance and selecting the maximum intensity value for each group, they can estimate the response of a perfectly homogeneous "virtual phantom."
- Algorithmic Calibration: This virtual data calibrates the actual patient measurements before image reconstruction using standard techniques like Tikhonov regularization within the NIRFAST software package.
Validation and Datasets
The method's efficacy was tested across three levels of complexity:
- Numerical Simulation: Monte Carlo simulations were performed on a 3D breast-sized domain with parabolic contours to test the mathematical foundations.
- Physical Phantoms: Experiments used Polyoxymethylene (POM) blocks with cylindrical inclusions filled with Intralipid and India ink to simulate the optical contrast of tumors.
- Clinical Trial: The system was tested on a 60-year-old patient with invasive micropapillary carcinoma, with results cross-validated against PET scans.
Key Results
The self-calibration method demonstrated high fidelity and clinical relevance:
- Accuracy: In simulations, the method achieved a relative error of only 4.43% compared to traditional physical phantom calibration.
- Contrast Recovery: Phantom experiments showed nearly identical absorption coefficient reconstruction between the self-calibration and conventional methods (a contrast ratio of 0.95:1).
- Clinical Success: In the human subject, the DOT reconstruction successfully localized a tumor region corresponding to high-uptake areas in PET scans. The system detected a 2.46x increase in total hemoglobin (HbT) in the malignant region, aligning with the known hypermetabolic characteristics of breast cancer.
Conclusion and Future Outlook
By replacing physical reference phantoms with a data-driven "virtual" alternative, this self-calibration method significantly streamlines the clinical application of breast DOT. While the technique currently requires high-density sensor arrays and assumes a relatively homogeneous background, it offers a robust path forward for practical, high-throughput optical imaging. Future work will focus on refining the method for more complex heterogeneous tissues and integrating it into multi-modal diagnostic suites to improve specificity in cancer detection.
Summary updated