Artificial intelligence (AI)-enabled denoising of ultra-low-dose chest computed tomography (CT) scans may facilitate key findings with pneumonia in immunocompromised individuals at a small fraction of normal CT radiation dosing.
For the prospective study, recently published in Radiology: Cardiothoracic Imaging, researchers compared normal-dose chest CT, ultra-low-dose CT (ULDCT) and denoised ULDCT for the detection of pneumonia in 54 immunocompromised adults (median age of 62).
The study authors found that denoised ULDCT improved the signal-to-noise ratio (SNR) by 20.16 percent in comparison to ULDCT (39.08 SNR vs. 31.2 SNR). Denoised ULDCT also demonstrated 10 percent and 12 percent better sensitivity rates than ULDCT for fungal infection (100 percent vs. 90 percent) and bacterial infection (100 percent vs. 88 percent) respectively, according to the researchers.
The study findings revealed that denoised ULDCT was more effective than ULDCT at ruling out pneumonia (100 percent vs. 90 percent). Researchers also noted that denoised ULDCT corrected two cases of false positives with ULDCT.
“Our artificial intelligence–based denoising technique significantly reduced the noise inherent in ULDCT and improved the identification of individuals who were immunocompromised without versus with pneumonia,” wrote lead study author Maximiliano Klug, M.D., who is affiliated with the Division of Diagnostic Imaging at the Chaim Sheba Medical Center in Ramat Gan, Israel, and colleagues.
The researchers determined that denoised ULDCT offered significantly enhanced detection of fine detail findings in the immunocompromised cohort in comparison to ULDCT. Specifically, the study authors pointed out greater than 13 percent accuracy in detecting tree-in-bud patterns (93 percent vs. 78-80 percent) as well as enhanced accuracy for interlobular septal thickening (78-83 percent vs. 61-67 percent).
“Denoised ULDCT improved detection of ground-glass opacities, tree-in-bud opacities, and inter- and intralobular septal thickening. Such fine detail identification may help support a particular diagnosis, such as an early inflammatory response, endobronchial infection, or subtle interstitial edema,” noted Klug and colleagues.
Three Key Takeaways
1. Improved pneumonia detection with AI-denoised ULDCT. AI-enhanced denoising of ultra-low-dose CT (ULDCT) improved sensitivity for detecting pneumonia in immunocompromised individuals and outperformed standard ULDCT in ruling out pneumonia and reducing false positives.
2. Enhanced identification of key imaging findings. Denoised ULDCT provided significantly better visualization of fine-detail abnormalities, including ground-glass opacities, tree-in-bud opacities, and interlobular septal thickening, aiding in more accurate diagnoses of infections and inflammatory responses.
3. Lower radiation exposure with high diagnostic value. Denoised ULDCT achieved similar diagnostic accuracy to normal-dose CT but with a substantially lower radiation dose (0.12 mSV vs. 6.15 mSV), making it a safer option for patients requiring frequent follow-up imaging.
Emphasizing a median effective radiation dose of 0.12 mSV for ULDCT in contrast to 6.15 mSV for conventional chest CT, the researchers maintained that denoised ULDCT could be a viable alternative for recurring CT follow-up exams in the immunocompromised population.
“Performing denoised ULDCT in lieu of normal-dose CT in young patients expected to undergo repetitive CT scans for infection evaluation should be considered to reduce cumulative radiation dose while preserving diagnostic accuracy,” added Klug and colleagues.
(Editor’s note: For related content, see “FDA Clears CT-Based AI Software for Enhanced Detection of Usual Interstitial Pneumonia,” “Study Assesses Lung CT-Based AI Models for Predicting Interstitial Lung Abnormality” and “MRI Study Shows Moderate to Severe Opacities Six Months After COVID-19 Pneumonia for One-Third Of Exams.”)
In regard to study limitations, the authors conceded the small cohort size and possible bias with subjective assessment of imaging.
