Purpose
Deep Learning Image Reconstruction (DLIR) promises significant improvements in computer tomography (CT) image quality and dose reduction [1]. Early evidence from phantom studies and selected clinical cases suggested that DLIR may significantly reduce image noise and enhance physiological features at lower doses than iterative image reconstruction methods (e.g. adaptive statistical iterative reconstruction, ASiR-V [2, 3]). This study evaluates DLIR on a broad range of abdominal scans from clinical practice.
Methods and materials
A comparison of image quality and dose was made between DLIR (high noise reduction level) and ASiR-V (factor 50%) for a CT protocol for the combined examination of the abdomen/pelvis. 54 patients (23/31 women/men, age 66 ± 10 years) have been examined in a CT Revolution (GE Healthcare).
The signal to noise ratio (SNR) and contrast to noise ratio (CNR) were calculated for the aorta, liver, spleen, kidney, pelvic bone and abdominal fat. The noise corresponds to the standard deviationσof HU values. The SNR was...
Results
Compared to ASiR-V, DLIR improves the visibility of physiological features at the same dose level (Fig. 1). The SNR was 5.08 ± 3.82 for DLIR and 2.93 ± 1.79 for ASiR-V. The SNR for DLIR is on average higher (Fig. 2): 2.00-fold (abdominal fat), 1.83-fold (spleen), 1.76-fold (kidney), 1.74-fold (liver), 1.63-fold (aorta) and1.11-fold (pelvic bone).
The CNR between organ and abdominal fat was 29.18 ± 23.39 for DLIR and 14.55 ± 11.59 for ASiR-V. The CNR for DLIR is on average 2-fold higher (Fig. 3).The...
Conclusion
At the same dose level, DLIR reduces noise and improves contrast. The improved image quality opens up new possibilities in dose optimization, high-resolution imaging and the examination of obese patients.
Personal information and conflict of interest
Andreas Heinrich, medical physicists
Department of Radiology, University Hospital Jena, Friedrich Schiller University, Germany
Am Klinikum 1
07747 Jena, Germany
e-mail:
[email protected]
A. Heinrich; Jena/DE - nothing to disclose N. D. Didier; Jena/DE - nothing to disclose M. Engler; Jena/DE - nothing to disclose U. K. M. Teichgräber; Jena/DE - nothing to disclose F. Güttler; Jena/DE - nothing to disclose
References
J. Hsieh, E. Liu, B. Nett, J. Tang, J.-B. Thibault, S. Sahney. “A new era of image reconstruction: TrueFidelityTM.”White Paper, GE Healthcare (2019).
J. Hsieh. “Adaptive statistical iterative reconstruction.”White Paper, GE Healthcare (2008).
Jiahua Fan, Meghan Yue, and Roman Melnyk, “Benefits of ASiR-V Reconstruction for Reducing Patient Radiation Dose and Preserving Diagnostic Quality in CT Exams.”White Paper, GE Healthcare (2014).