A data-driven maximum likelihood classification for nanoparticle agent identification in photon-counting CT.

Area of Science:

• Medical Imaging
• Nanotechnology
• Data Science

Background:

• Nanoparticle agents combined with targeting factors offer potential for specific CT imaging and improved clinical diagnosis.
• Photon-counting detectors (PCD) possess energy sensitivity, enabling exploitation of nanoparticle K-edges for agent identification within the clinical X-ray energy range.

Purpose of the Study:

• To propose and validate a novel data-driven approach for identifying nanoparticle agents using PCD measurements.
• To assess the method's robustness, particularly concerning dose reduction and varying agent concentrations.

Main Methods:

• Generation of training data from PCD measurements of calibration phantoms (with and without nanoparticle agents).
• Calculation of normalized log-likelihood sinograms using the K-nearest neighbors (KNN) estimator for each class.
• Backprojection of sinograms and comparison of images for agent identification, mathematically proven equivalent to maximum likelihood classification.

Main Results:

• The proposed method successfully identifies nanoparticle agents using PCD data.
• Demonstrated robustness of the approach with gold nanoparticles as K-edge contrast media, even with dose reduction.
• Effective identification of targets with varying concentrations of agents and resilience to background noise.

Conclusions:

• The developed data-driven method provides a reliable means for nanoparticle agent identification in CT imaging.
• This technique holds significant potential for enhancing the accuracy and safety of clinical diagnosis through improved imaging.
• The method's effectiveness under dose reduction scenarios highlights its clinical applicability and efficiency.