Deep learning reconstruction algorithm and high-concentration contrast medium: feasibility of a double-low protocol in coronary computed tomography angiography

Abstract

Objective: To evaluate radiation dose and image quality of a double-low CCTA protocol reconstructed utilizing high-strength deep learning image reconstructions (DLIR-H) compared to standard adaptive statistical iterative reconstruction (ASiR-V) protocol in non-obese patients.

Materials and methods: From June to October 2022, consecutive patients, undergoing clinically indicated CCTA, with BMI < 30 kg/m² were prospectively included and randomly assigned into three groups: group A (100 kVp, ASiR-V 50%, iodine delivery rate [IDR] = 1.8 g/s), group B (80 kVp, DLIR-H, IDR = 1.4 g/s), and group C (80 kVp, DLIR-H, IDR = 1.2 g/s). High-concentration contrast medium was administered. Image quality analysis was evaluated by two radiologists. Radiation and contrast dose, and objective and subjective image quality were compared across the three groups.

Results: The final population consisted of 255 patients (64 ± 10 years, 161 men), 85 per group. Group B yielded 42% radiation dose reduction (2.36 ± 0.9 mSv) compared to group A (4.07 ± 1.2 mSv; p < 0.001) and achieved a higher signal-to-noise ratio (30.5 ± 11.5), contrast-to-noise-ratio (27.8 ± 11), and subjective image quality (Likert scale score: 4, interquartile range: 3-4) compared to group A and group C (all p ≤ 0.001). Contrast medium dose in group C (44.8 ± 4.4 mL) was lower than group A (57.7 ± 6.2 mL) and B (50.4 ± 4.3 mL), all the comparisons were statistically different (all p < 0.001).

Conclusion: DLIR-H combined with 80-kVp CCTA with an IDR 1.4 significantly reduces radiation and contrast medium exposure while improving image quality compared to conventional 100-kVp with 1.8 IDR protocol in non-obese patients.

Clinical relevance statement: Low radiation and low contrast medium dose coronary CT angiography protocol is feasible with high-strength deep learning reconstruction and high-concentration contrast medium without compromising image quality.

Key points: Minimizing the radiation and contrast medium dose while maintaining CT image quality is highly desirable. High-strength deep learning iterative reconstruction protocol yielded 42% radiation dose reduction compared to conventional protocol. "Double-low" coronary CTA is feasible with high-strength deep learning reconstruction without compromising image quality in non-obese patients.

Keywords: Computed tomography angiography; Contrast media; Deep learning; Image processing; Radiation dosage.

Fig. 1

Flow diagram of patient recruitment. BMI, body mass index; DLIR-H, high strength deep learning iterative reconstruction; IDR, iodine delivery rate; ASIR, adaptive statistical iterative reconstruction

Fig. 2

Box-and-whisker plots for quantitative image quality show the distribution of coronary arteries attenuation values (A), SNR (B), and CNR (C) of groups A, B, and C. Boxes represent the middle 50% of the data, solid lines represent the median, and whiskers represent minimum and maximum values. Group B yielded significantly higher attenuation values, SNR, and CNR (all p < 0.001)

Fig. 3

Curved multiplanar reformations depicting LAD arteries of a 53-year-old man assigned to group A (A), a 69-year-old woman assigned to group B (B), and a 63-year-old woman assigned to group C (C), respectively. All studies were deemed diagnostic. Group B (score: 4, interquartile range: 3–4) outperformed both group C (score: 3, interquartile range: 3–3) and group A (score: 3, interquartile range: 3–4; all p < 0.001); no statistically significant differences have been found between group A and C (p = 0.338)