Deep Learning-Based Augmented Contrast-Enhancement and Denoising for Reduced-Iodine and Low-Radiation 70-kVp Cerebral...

Area of Science:

• Radiology
• Medical Imaging
• Computed Tomography

Background:

• Cerebral computed tomography angiography (CTA) traditionally uses higher radiation doses and iodine concentrations.
• Optimizing CTA protocols for reduced radiation and iodine load is crucial for patient safety and minimizing contrast-related risks.
• Advanced image processing techniques are being explored to maintain or improve image quality under reduced radiation and iodine conditions.

Purpose of the Study:

• To assess the feasibility of cerebral CTA using a low radiation (70 kVp) and reduced iodine protocol.
• To evaluate the impact of deep learning-based augmented contrast enhancement (DL-ACE) and denoising (DL-DN) algorithms on CTA quality in this reduced-protocol setting.

Main Methods:

• A prospective study involving 47 healthy volunteers assigned to three CTA groups with varying kVp, iodine concentration, and volume.
• Group A: 100 kVp, 40 mL of 350 mgI/mL iodine.
• Groups B & C: 70 kVp with reduced iodine (270 mgI/mL) and volume (40 mL and 28 mL ultralow iodine).
• Images were reconstructed using filtered back projection (FBP), with DL-ACE and DL-DN applied to Groups B and C.
• Comparison of arterial attenuation, image noise, contrast-to-noise ratio (CNR), and subjective image quality.

Main Results:

• Groups B and C achieved a 23.7% reduction in radiation dose compared to Group A.
• Without deep learning, 70 kVp protocols showed higher arterial attenuation but also increased noise, leading to similar CNR.
• Application of DL-ACE and DL-DN in Groups B and C significantly increased arterial attenuation (by 45.4%), decreased noise (by 34.5%), and improved CNR and subjective image quality compared to Group A.

Conclusions:

• Cerebral CTA at 70 kVp with ultralow iodine enhances arterial attenuation but increases image noise compared to 100 kVp.
• DL-ACE and DL-DN algorithms are effective in mitigating increased noise and enhancing attenuation, leading to superior image quality.
• These findings support the feasibility of low-radiation, reduced-iodine cerebral CTA protocols enhanced by deep learning techniques.