Photon-counting computed tomography thermometry via material decomposition and machine learning.

Area of Science:

• Medical Physics
• Biomedical Imaging
• Computational Biology

Background:

• Thermal ablation uses minimally invasive heating to eliminate tumors.
• Real-time 3D temperature visualization is crucial for accurate targeting and minimizing collateral damage.
• Current computed tomography (CT) thermometry relies on energy-integrated CT and linear attenuation-temperature relationships, limiting accuracy.

Purpose of the Study:

• To develop a novel photon-counting CT approach for accurate 3D temperature prediction.
• To utilize material decomposition and neural networks for enhanced thermal property analysis.
• To overcome limitations of traditional CT thermometry in non-linear thermal scenarios.

Main Methods:

• Developed a novel approach using photon-counting CT for material decomposition.
• Employed a neural network to predict temperature based on material thermal characteristics and spectral tomographic measurements.
• Conducted a feasibility study using distilled water, 50 mmol/L CaCl2, and 600 mmol/L CaCl2 as base materials, measuring attenuation in four energy bins at various temperatures.

Main Results:

• The neural network achieved a mean absolute error of 3.97°C on 300 mmol/L CaCl2.
• The model demonstrated a mean absolute error of 1.80°C on a milk-based protein shake.
• Experimental results show promise for handling non-linear thermal properties in diverse materials.

Conclusions:

• The developed photon-counting CT thermometry approach shows significant potential for accurate real-time 3D temperature visualization.
• This method offers improved accuracy over traditional CT thermometry, especially for materials with non-linear thermal properties.
• The findings suggest a promising new tool for enhancing the precision and safety of thermal ablation procedures.