Deep learning algorithms empower personalised low-dose CT scanning: Application value in liver space-occupying lesions

Objective This study aimed to evaluate the clinical utility of a personalised low-radiation-dose, low-iodinated-contrast-agent, and low-injection-rate CT imaging protocol utilising deep learning-based reconstruction algorithms for hepatic space-occupying lesions.Methods Eighty patients with suspected or confirmed hepatic space-occupying lesions undergoing upper abdominal contrast-enhanced CT were prospectively enrolled. The conventional group (Group A, n=40) employed the following parameters: tube voltage 120 kV, automatic tube current modulation (O-Dose technology), iterative reconstruction algorithm (CV) with 60% weighting, and injection protocol: Iodinated contrast agent concentration 370 mgI/mL, injection rate 3.8 ml/s, total contrast volume 85 ml; The ‘triple-low’ group was designated as Group B (n=40): tube voltage was individually adjusted based on the patient's abdominal fat area (TFA), tube current utilised O-Dose technology, and four image sets were reconstructed using CV 60% and deep learning reconstruction algorithms (CI) with 40%, 60%, and 80% weighting, designated as Groups B1 to B4. Injection protocol: Iodinated contrast agent concentration 300 mgI/mL or 370 mgI/mL, with injection rate and contrast volume dynamically adjusted according to patient lean body mass. Measure CT values and standard deviation (SD) of hepatic parenchyma at the portal level in the arterial phase, left paravertebral muscles, and hepatic lesions. Calculate signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and figure of merit (FOM). Overall image quality, lesion display, artefact severity, and diagnostic confidence were assessed and compared across groups. Effective radiation dose (ED), body-specific dose estimate (SSDE), patient iodine uptake, and iodine infusion rate were calculated for Groups A and B.Results No statistically significant differences existed in general clinical characteristics between Groups A and B (P > 0.05). Compared with Group A, Group B exhibited reductions in ED, SSDE, patient iodine uptake, and injection flow rate of 43.28%, 27.56%, 20.67%, and 40.79%, respectively (P < 0.05). Group B3 demonstrated higher liver parenchymal SNR, lesion SNR, and CNR than Group A (P < 0.05). Inter-observer agreement between two radiologists demonstrated good consistency (Kappa values: 0.708–0.954, P < 0.05). Within Group B, subjective scores were highest in B3, which also exceeded those of Group A (P < 0.05).Conclusion The CI algorithm-based individualised imaging protocol for enhanced CT of hepatic lesions significantly reduces radiation dose, patient iodine intake, and iodine infusion rate while maintaining image quality, thereby enhancing patient safety. This approach demonstrates clinical applicability, with 60% CI identified as the optimal reconstruction weighting. Trial registration Chinese Clinical Trial Registry ，PJ-KS-KY-2025-590