Skip to main content
Diagnostic imaging

Diagnostic imaging

Deep learning can decrease radiation dose in paediatric CT scans

29 Mar 2022
Contrast-enhanced 80 kVp CT images
Reducing the scan dose Contrast-enhanced low-dose CT images of a 4-year-old boy, reconstructed using hybrid iterative reconstruction (HIR), model-based iterative reconstruction (MBIR) and deep learning-based reconstruction (DLR). (Courtesy: Yasunori Nagayama)

Deep learning-based reconstruction (DLR), an emerging technique for reconstructing CT images, uses a convolutional neural network to produce low-noise, high-quality images in short times. Researchers in Japan have now demonstrated that DLR can enable substantial dose reduction in paediatric CT exams with the same, or even improved, image quality compared with the use of iterative reconstruction algorithms.

Writing in the American Journal of Roentgenology, the team reports that using DLR for low-tube-voltage exams reduced the image noise without degrading noise texture and image sharpness relative to hybrid iterative reconstruction (HIR) and model-based iterative reconstruction (MBIR).

Because children are more sensitive to ionizing radiation than adults, using the lowest possible radiation dose to obtain diagnostic quality images is the goal of every radiology professional performing paediatric CT. One effective technique for dose reduction in paediatric contrast-enhanced CT combines decreased tube voltage (80 kVp, for example, as opposed to the standard 120 kVp) with iterative reconstruction.

Decreasing the tube voltage, however, increases image noise and can impair detection of low-contrast objects, especially when using reduced slice thicknesses to evaluate small anatomic structures in children, explains principal investigator Yasunori Nagayama of Kumamoto University. And although HIR and MBIR can reduce noise and artefacts, they have limited ability to preserve noise texture, low-contrast spatial resolution and low-contrast object detectability at considerably reduced doses.

To assess and compare the image quality achieved by HIR, MBIR and DLR, the researchers retrospectively analysed the scans of 65 children aged six years and under who underwent contrast-enhanced abdominal CT, 31 with a standard protocol and 34 with a lower-dose protocol. All CT exams were performed using a tube voltage of 80 kVp and automated tube current modulation of mild and standard dose reduction strength, for the standard and lower-dose protocols, respectively.

The team used the AiCE (Advanced intelligent Clear-IQ Engine) Body-sharp algorithm for DLR. All image reconstructions used a noise reduction level of “standard” as well as 1-mm slice thickness and increment.

Two radiologists independently evaluated the noise magnitude, noise texture, streak artefact, edge sharpness and overall quality of each reconstructed image, using a subjective four-point scale. The research team also quantified the image noise, signal-to-noise ratio and contrast-to-noise ratio for all scans. Calculating the size-specific dose estimate (SSDE) for both protocols revealed that SSDE was 54% lower in the lower-dose group than the standard group (1.9±0.4 versus 4.0±1.0 mGy).

The researchers also performed a phantom experiment, using a 20-cm cylindrical phantom, to assess image quality at a wider range of dose settings. They employed the same CT scanner and image parameters as for the clinical patients, using eight fixed tube currents to achieve doses relevant to clinical paediatric CT. Again, they reconstructed the data using HIR, MBIR and DLR with a standard noise reduction level and 1-mm slice thickness.

The team reports that low-dose DLR images had significantly higher subjective scores for noise magnitude, noise texture, edge sharpness and overall quality than standard HIR, low-dose HIR and low-dose MBIR images. DLR also outperformed HIR and MBIR in the phantom analysis.

“This clinical and phantom investigation indicated that, compared with iterative reconstruction algorithms, DLR reduced radiation dose by approximately 50% while preserving or even improving image quality and task-based object detectability for contrast-enhanced 80-kVp CT in young children,” the researchers conclude. “The findings may be applied to achieve substantial radiation dose reductions in paediatric CT in comparison with current IR techniques.”

Copyright © 2024 by IOP Publishing Ltd and individual contributors