A Physics-ASIC Architecture-Driven Deep Learning Photon-Counting Detector Model Under Limited Data.

Area of Science:

• Medical Imaging
• Detector Physics
• Artificial Intelligence

Background:

• Photon-counting computed tomography (PCCT) offers advanced imaging capabilities.
• Accurate modeling of photon-counting detectors (PCDs) is crucial but challenging due to complex, nonlinear responses and limited calibration data.
• Current limitations hinder the widespread adoption of PCCT technology.

Purpose of the Study:

• To develop a novel deep learning detector model for PCDs.
• To accurately capture both sensor and ASIC responses within PCDs.
• To address the challenge of modeling complex PCDs with limited calibration data.

Main Methods:

• Introduction of a physics-ASIC architecture-driven deep learning model.
• The model integrates sensor and application-specific integrated circuit (ASIC) responses.
• Validation using experimental data with limited calibration sets.

Main Results:

• Demonstrated exceptional accuracy and robustness of the deep learning model.
• Achieved significant reduction in calibration errors.
• Obtained reasonable estimation of physics-ASIC parameters.
• Generated high-quality, high-accuracy material decomposition images.

Conclusions:

• The proposed deep learning model effectively addresses the challenges in PCCT detector modeling.
• This approach enhances the accuracy and reliability of material decomposition.
• The findings pave the way for broader accessibility and application of PCCT.