Dual-energy cone-beam CT with a flat-panel detector: effect of reconstruction algorithm on material classification

Area of Science:

• Medical Imaging
• Radiology
• Cone-Beam Computed Tomography (CBCT)

Background:

• Dual-energy (DE) imaging capabilities in Cone-beam CT (CBCT) with flat-panel detectors (FPDs) show promise for applications like breast and musculoskeletal imaging.
• Accurate material classification is crucial for enhancing diagnostic information in DE CBCT.

Purpose of the Study:

• To investigate the material classification accuracy of DE CBCT using filtered backprojection (FBP) and penalized likelihood (PL) reconstruction methods.
• To optimize contrast-enhanced DE CBCT of joints by analyzing the impact of dose, material concentration, and detail size.

Main Methods:

• Phantoms with calcium and iodine inserts, and a cadaveric knee with iodine injection, were imaged using a CBCT system with specific low (LE) and high (HE) energy beams.
• DE classification was performed using a nearest distance classifier, with differential filtering (FBP) or regularization (PL) applied to LE and HE data to account for noise differences.
• Reconstruction methods included FBP, quadratic penalized likelihood (PLQ), and total-variation penalized likelihood (PLTV), with varying total doses (3.1–15.6 mGy).

Main Results:

• Classification accuracy was influenced by noise in both LE and HE images, with HE noise impacting contrast inserts and LE noise affecting surrounding water.
• Optimized DE CBCT reconstructions achieved high accuracy (up to 98% with PLTV at 15.6 mGy) for larger inserts (28.4 mm).
• Edge-preserving PLTV reconstruction significantly improved the minimum detectable diameter (down to 3 mm) compared to FBP and PLQ at higher doses.

Conclusions:

• High material classification accuracy (>90%) for iodine and bone in DE CBCT is achievable at doses below 10 mGy for concentrations >5 mg/ml and detail sizes ~20 mm.
• Optimal performance relies on reconstruction parameter selection, often favoring stronger smoothing of HE data and using task-specific penalties like PLTV.
• DE CBCT successfully demonstrated discrimination of iodine and bone in a cadaveric knee model.