Related Experiment Video
Updated: Jul 16, 2025
Novel In Vivo Micro-Computed Tomography Imaging Techniques for Assessing the Progression of Non-Alcoholic Fatty Liver Disease
Published on: March 24, 2023
Low-dose liver CT: image quality and diagnostic accuracy of deep learning image reconstruction algorithm
Damiano Caruso1, Domenico De Santis1, Antonella Del Gaudio1
1Department of Medical-Surgical Sciences and Translational Medicine, Radiology Unit, Sant'Andrea University Hospital, Sapienza University of Rome, Via Di Grottarossa, 1035-1039, 00189, Rome, Italy.
Deep learning image reconstruction (DLIR) offers superior image quality and diagnostic accuracy for abdominal CT scans compared to adaptive statistical iterative reconstruction (ASiR-V). DLIR particularly improves detection of small hypovascular liver lesions.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Healthcare
- Abdominal CT Imaging
Background:
- Iterative reconstruction algorithms are standard in abdominal CT, but their impact on image quality and diagnostic accuracy for hypovascular liver lesions requires further evaluation.
- Adaptive statistical iterative reconstruction (ASiR-V) is a widely used iterative reconstruction technique.
- Deep learning image reconstruction (DLIR) presents a novel approach to image reconstruction with potential benefits for image quality and lesion detection.
Purpose of the Study:
- To conduct a within-subject analysis of abdominal CT image quality reconstructed with DLIR.
- To compare the diagnostic accuracy of DLIR against the standard ASiR-V algorithm.
- To evaluate the performance of DLIR across different intensity levels for hypovascular liver lesions.
Main Methods:
- Prospective enrollment of oncologic patients undergoing contrast-enhanced abdominal CT.
- Image reconstruction using DLIR (high, medium, low intensity) and ASiR-V (10-100% strength).
- Radiologist characterization of lesions and reader assessment of diagnostic accuracy, SNR, CNR, FOM, and subjective image quality using a Likert scale.
Main Results:
- DLIR_H demonstrated comparable SNR and CNR to ASiR-V 100%.
- DLIR_M achieved the highest subjective image quality and a significant increase in FOM.
- DLIR_M improved detection of lesions ≤0.5 cm (32/33) compared to ASiR-V 50% (26/33), with overall lesion accuracy of 93.8% for DLIR versus 87.7% for ASiR-V 50%.
Conclusions:
- DLIR provides superior image quality and enhanced diagnostic accuracy for hypovascular liver lesions compared to ASiR-V.
- The medium-strength DLIR algorithm shows particular efficacy in detecting small hypovascular liver lesions (≤0.5 cm).
- DLIR can be safely integrated into routine abdominal CT protocols as a replacement for iterative reconstruction techniques.
