Adding Deep Learning to Ultra-Low Dose CT Drastically Reduces Exposure for Emphysema Patients

Whitney J. Palmer

Cumulative radiation dose exposure is a concern for this group.

Pairing deep-learning noise reduction (DLNR) with ultra-low dose CT (ULDCT) for the quantification of emphysema can reduce a patient’s dose expose by more than 80 percent compared to standard-dose CT (SDCT).

Although preventable, emphysema, one of the diseases of chronic obstructive pulmonary disorder (COPD), is incurable once it develops, and more than 3 million American adults have been diagnosed with the condition. CT scans are typically used to monitor COPD progression, so a patient’s lifetime cumulative dose exposure from SDCT could be considerable.

In a study published March 10 in the European Journal of Radiology, investigators from the University Medical Center Groningen in The Netherlands showed it is possible to drastically reduce radiation dose – a 84-percent drop – with only a slight underestimation of emphysema.

Related Content: Ultra-Low Dose CT vs Standard-Dose to Quantify Emphysema

“The decrease in detail may reduce the differentiation between emphysema and healthy lung tissue by blurring the image,” said the team led by Rozemarijn Vliegenthart, a radiologist and member of the faculty of medical sciences. “It is likely that there is an optimal setting where a substantial part of the noise is removed, but structural details are still mostly visible in the image, allowing accurate quantification of emphysema.”

To determine how well ULDCT performed, Vliegenthart’s team recruited 49 patients with COPD who underwent both the ultra-low dose scan, as well as SDCT between February 2018 and June 2018. Scans lasted 30 minutes and were conducted on the same day.

Investigators used a standard protocol on the high-resolution SDCT scans. ULDCT scans were acquired with automatic exposure control that ensured, even with the very low dose, that images were adequate and uniform. Both scans were reconstructed with filtered backprojection (FBP) and soft kernel, and the researchers also applied advanced modeled iterative reconstruction (ADMIRE) levels 1, 3, and 5, and DLNR levels 1, 3, 5, and 9. They quantified emphysema as Low Attenuation Value percentage (LAV%).

At assessment, 51 percent of patients exhibited emphysema of, at least, moderate severity. Based on the team’s comparison and analysis, the median dose-length product (DLP) of the ULDCT was 16.6 – 84 percent lower than the average SDCT radiation dose. In addition, the median extent of emphysema was 18.6 percent with ULDCT-FBP versus 15.4 percent with SDCT.

Their results also pointed to optimal settings for variability of emphysema quantification – ADMIRE 3 and DLNR 3 produced a reduction of 24 percent and 27 percent, respectively, with a slight underestimation of emphysema extent of -1.5 percent and -2.9 percent each.

The team did note one particular strength of their study results – DLNR software can be applied to CT scans in a vendor-neutral way even long after the scan has been acquired. This capability adds to the generalizability of their results, they added.

Still, there is more work to be done, they concluded.

“Subsequent investigations should determine if a simple baseline correction is sufficient to correct for the systematic bias and reliably determine the level of parenchymal destruction,” they said. “Future research is needed to assess if the scans are sufficiently accurate and detailed, so that no relevant structural information required for visual assessment is lost.”

For more coverage based on industry expert insights and research, subscribe to the Diagnostic Imaging e-Newsletter here.