A novel method of micro-tomography geometric angle calibration with random phantom

Area of Science:

• Medical Imaging
• Machine Learning
• Computational Science

Background:

• Micro-tomography systems are increasingly vital in medical applications like surgery and radiology.
• Image artifacts from CT construction errors can obscure actual pathology in reconstructed volumes.
• Accurate geometry calibration is crucial for reliable micro-tomography data.

Purpose of the Study:

• To develop and assess a machine learning approach for micro-tomography geometry angle calibration.
• To address CT construction errors that lead to image artifacts.
• To enhance the diagnostic and treatment planning capabilities of micro-tomography.

Main Methods:

• A Kernel Ridge Regression algorithm was developed for micro-tomography geometry estimation.
• A specialized phantom with steel ball bearings was used for calibration.
• Projection images and gantry angle information were utilized to calibrate in-plane and out-plane detector rotations.

Main Results:

• The Kernel Ridge Regression algorithm successfully calibrated micro-tomography geometry parameters.
• Computer simulations validated the accuracy of the geometry parameter calibration.
• The method demonstrated potential for calibrating other CT construction parameters.

Conclusions:

• Machine learning, specifically Kernel Ridge Regression, offers a feasible and accurate method for micro-tomography geometry calibration.
• This approach can reduce image artifacts and improve the reliability of micro-tomography data.
• The developed technique has broad applicability for enhancing CT system calibration.