How AI reduces radiation dose, but not quality, of key imaging findings

Deep learning-based reconstruction (DLR) can reduce the radiation dose associated with low-dose chest and abdominal CT scans without sacrificing image quality, according to a new study published in the American Journal of Roentgenology. The authors compared DLR with traditional iterative reconstruction (IR) and filtered back-projection (FBP) reconstruction techniques.

“Although IR methods have gained wide acceptance over FBP, multiple studies have reported that they cannot enable submillisievert CT for routine indications such as a single-phase routine abdominal CT,” wrote lead author Ramandeep Singh, Massachusetts General Hospital in Boston, and colleagues. “Previous study designs have combined quantitative and qualitative evaluation for assessment of different image reconstruction techniques. Recently, DLR methods using deep convolutional neural networks (DCNNs) have been proposed to enable dose reduction while maintaining the diagnostic performance of CT.”

To put DLR to the test, the researchers explored data from 59 adult patients who underwent routine chest and abdominopelvic CT exams. All patients agreed to receive both low-dose CT (LDCT) and standard-dose CT. While the standard-dose exams were reconstructed using IR, the low-dose exams were reconstructed using IR, FBP and DLR.

Overall, low-dose exams were associated with a lower mean volume CT dose index (2.1 ± 0.8 mGy) and dose-length product (49 ± 13mGy·cm) than the standard-dose exams. The low-dose exams reconstructed with DLR were rated as “acceptable for interpretation” for 97% of abdominal studies and 95%-100% of chest studies.

Also, low-dose exams reconstructed using IR and FBP DLR had “inferior image quality” compared to standard-dose exams. The performance of DLR, however, was more comparable to standard-dose exams. In addition, DLR achieved “substantially better” image noise and signal-to-noise ratios than standard-dose exams for both chest and abdominal exams.

“The image quality improvements with DLR relative to both FBP and IR techniques should allow users to reduce radiation dose when using DLR with routine chest and abdominal CT examinations,” the authors wrote. “Our study highlights the need for additional studies with a broad spectrum of abnormalities and protocols in the chest (such as nonpulmonary findings) and the abdomen (such as for renal colic evaluation).”