Gaussian process classification of superparamagnetic relaxometry data: Phantom study.

Area of Science:

• Biomedical Engineering
• Medical Imaging
• Machine Learning

Background:

• Superparamagnetic relaxometry (SPMR) is an emerging technology for early cancer detection.
• SPMR relies on detecting magnetic fields from superparamagnetic iron oxide nanoparticles (SPIONs) bound to tumors.
• Current image reconstruction methods are computationally intensive and sensitive to noise.

Purpose of the Study:

• To evaluate a data-driven machine learning technique for detecting weak SPION signals indicative of tumors.
• To compare the performance of Gaussian process (GP) classification against traditional image reconstruction.
• To identify scenarios where machine learning can replace complex reconstruction techniques.

Main Methods:

• Utilized Gaussian process (GP) classification and a physics-based image reconstruction method.
• Employed both in silico simulations based on mouse cancer models and phantom experiments.
• Investigated performance under varying measurement noise, SPION distribution, and concentration levels.

Main Results:

• GP classification achieved high sensitivity in silico, outperforming image reconstruction.
• Phantom studies successfully detected surrogate tumors with low SPION concentrations (5% and 7.3%).
• Accuracies of 87.5% and 96.4% were achieved in phantom studies.

Conclusions:

• The GP framework demonstrates effective classification for SPMR data.
• GP classification offers a viable alternative to image reconstruction for binary classification tasks in SPMR.
• This data-driven approach shows potential for improving early cancer detection via SPMR.