Geometry adaptive projection-domain deep scatter estimation for multi-source semi-stationary cone-beam computed...

Area of Science:

• Medical Imaging
• Computational Imaging
• Radiological Physics

Background:

• Stationary or semi-stationary CBCT (sCBCT) offers potential for compact imaging systems.
• sCBCT geometries face challenges with complex X-ray scatter, hindering traditional scatter estimation methods.
• Reconstruction artifacts in sCBCT limit the effectiveness of volume-based scatter estimators.

Purpose of the Study:

• Propose adaptive deep scatter estimation (ADSE), an adaptive projection-domain technique for sCBCT.
• Overcome limitations of existing projection- and volume-domain scatter estimators in sCBCT.
• Improve the applicability of scatter estimation in sCBCT configurations.

Main Methods:

• Transform sCBCT projections to a view-invariant surrogate geometry.
• Apply an iterative, CNN-based scatter estimator and fluence weighting in the surrogate geometry.
• Obtain final sCBCT scatter estimates via inverse transformations and weighting.

Main Results:

• ADSE achieved a 3.88% MAPE in projection-domain scatter magnitude for non-truncated projections, outperforming iMC (4.42%) and gDSE (5.13%).
• In physical phantom experiments, ADSE recovered 48.67% of contrast and 25.03% of CNR, significantly improving upon iMC and gDSE.
• ADSE reduced cupping artifacts by 79% and CT number non-uniformity by 71%.

Conclusions:

• ADSE effectively addresses scatter estimation challenges in complex sCBCT geometries, including truncation and under-sampling.
• ADSE demonstrates superior performance over geometry-aware gDSE and iterative Monte Carlo (iMC) methods.
• The findings support the feasibility of scatter compensation in sCBCT using tailored geometrical warping operators.