Time to Scrutinize and Revise the Fine Print of Lung Cancer Screening Using Low-Dose CT: Seeking Greater Confidence in Cancer Detectability
2022; Radiological Society of North America; Volume: 303; Issue: 1 Linguagem: Inglês
10.1148/radiol.213084
ISSN1527-1315
Autores Tópico(s)Lung Cancer Diagnosis and Treatment
ResumoHomeRadiologyVol. 303, No. 1 PreviousNext Reviews and CommentaryFree AccessEditorialTime to Scrutinize and Revise the Fine Print of Lung Cancer Screening Using Low-Dose CT: Seeking Greater Confidence in Cancer DetectabilityHo Yun Lee Ho Yun Lee Author AffiliationsFrom the Department of Radiology and Center for Imaging Science, Samsung Medical Center, Sungkyunkwan University School of Medicine, 81 Irwon-Ro, Gangnam-Gu, Seoul 06351, Korea; and Department of Health Sciences and Technology, Samsung Advanced Institute for Health Sciences & Technology, Sungkyunkwan University, Seoul, Korea.Address correspondence to the author (e-mail: [email protected]).Ho Yun Lee Published Online:Jan 18 2022https://doi.org/10.1148/radiol.213084MoreSectionsPDF ToolsImage ViewerAdd to favoritesCiteTrack CitationsPermissionsReprints ShareShare onFacebookTwitterLinked In See also the article by Jiang et al in this issue.Dr Lee is an associate professor in the thoracic imaging section of the Department of Radiology at Sungkyunkwan University School of Medicine and Samsung Medical Center. Her research interests are thoracic oncologic imaging and radiomics, focusing on quantitative analysis and integration with genomics for prognostic stratification or therapy response evaluation. She is currently a chair of the committee on planning of the Korean Workshop for Pulmonary Functional Imaging and a member of the Fleischner Society.Download as PowerPointOpen in Image Viewer Since 2011, annual screenings using low-dose CT have been performed worldwide because they can reduce lung cancer mortality by 20% (1). However, lung cancer screening using low-dose CT has drawbacks: It can cause unnecessary invasive procedures or complications because it has a high false-positive rate associated with malignancy (23%), and the cumulative burden of radiation exposure from low-dose CT as compared with conventional radiography every year limits its implementation on a wider scale (2). There has also been concern over the low-dose CT technique, which may cause increased image noise and thus degrade image quality.With the recent rapid advancement of CT technology, tolerable-quality CT scans have been readily available at an exceptionally small quantity of radiation exposure, with an effective dose of less than 1 mSv (sub-millisievert level, expressed as ultra–low-dose [ULD] CT) (3), resulting in much more deteriorating image noise. To remedy this shortcoming, since the first commercially available iterative reconstruction methods appeared in the late 2000s (4), a range of noise reduction techniques has been introduced. Examples include statistical and model-based iterative reconstruction, as well as deep learning reconstructions (5). With the increasing demand for lower CT doses in recent years, the evaluation of noise reduction techniques has become progressively more relevant and widely studied.An iterative reconstruction algorithm is a procedure that generates an image through multiple processes, including iterative filters and back and forward projections using raw data. Iterative reconstruction algorithms were introduced clinically in 2009 to obtain better image quality with a lower radiation dose compared with a single reconstructed filtered back projection (FBP) (5). To date, various techniques—such as statistical, adaptive statistical, and fully model-based iterative reconstruction—have been developed by various vendors through a combination of noise, object, and physics modeling (6). However, conventional iterative reconstruction has two major limitations: reconstruction time and image texture. As many techniques are used to reduce image noise, the reconstruction time becomes longer, and unnatural textures are produced despite numeric improvements in parameters reflecting image noise, such as signal-to-noise ratio and contrast-to-noise ratio (7,8). Unlike FBP reconstruction, which is a linear process, these noise reduction techniques are nonlinear processes and can therefore generate output images with unique characteristics and appearance. Consequently, image denoising algorithms using artificial neural networks, termed deep learning image reconstruction (DLIR), have been applied to CT image reconstruction to overcome the drawbacks of iterative reconstruction while achieving good image quality (8,9). These studies showed that DLIR can reduce noise and improve spatial resolution but was limited by phantom imaging or retrospective study designs.In this issue of Radiology, Jiang and colleagues (10) report a well-designed study that investigates the performance of DLIR with regard to image quality and nodule assessment at ULD CT as low as 0.07 mSv or 0.14 mSv while comparing with other reconstruction methods, such as FBP or adaptive statistical iterative reconstruction-V (ASIR-V). Even in their prospective study of 203 participants with 1066 nodules, DLIR significantly reduced the background image noise in chest radiography–equivalent ULD CT and even reached a normal-dose CT level, compared with ASIR-V (mean air background noise ± standard deviation, 23 HU ± 4 vs 29 HU ± 4; P < .001). Moreover, DLIR significantly improved the detection rate of lung nodules from that of FBP (75.8% vs 62.5%) and also had a higher detection rate than ASIR-V did (73.3%) (P < .001). In terms of the recent deep learning approach in various imaging fields, the authors’ focus on improving the image quality particularly in LDCT for screening is very timely and has an obvious clinical potential.The strength of this study is its coverage of both quantitative and qualitative aspects of nodule evaluation. In terms of quantitative issues with the assessment of detected nodules, the authors compared DLIR with FBP and ASIR-V for diameter and volume of nodules according to nodule solidity or density, including subsolid, solid, or calcified lesions. The authors found that DLIR slightly underestimated the long diameter and volume of subsolid nodules but overestimated that of solid or calcified nodules. As the authors emphasize, the segmentation algorithm works to different levels according to different nodule densities due to the partial volume effect when defining the interface between high-attenuating nodules and low-attenuating lung parenchyma. While the transition zone around solid or calcified nodules leads to overestimation, part-solid or nonsolid nodules have blurred margins and low attenuation and are therefore indistinguishable from lung parenchyma, which reduces the accuracy of nodule segmentation, resulting in underestimation. These observations provide pragmatic information in real-world practice. This is particularly evident when applying the Lung CT Screening Reporting and Data System, or Lung-RADS, based on size criteria. Meanwhile, the authors also deal with the possibility of different interpretation according to noise reduction techniques in terms of semantic or qualitative nodule characteristics. To be more specific, they compared the diagnostic performance of malignancy-related imaging features, including lobulated shapes, spiculated margins, pleural tags, and air bronchograms. When allowing for the influence of ULD CT on the diagnostic evidence of malignant nodules, this approach is relevant and practical, given that there are sometimes morphologic characteristics more important than size by which to judge malignant or benign nodules. As a result, DLIR ensured the nodule measurement accuracy and displayed more malignancy-related imaging features of nodules than ASIR-V did.The consistency of detectable nodules and undetectable nodules of the three different noise reduction techniques was not addressed in the study by Jiang et al (10). Considering that noise reduction logic works differently for each technique, we are uncertain if the three techniques of FBP, ASIR-V, and DLIR have similar properties for detectability, to greater or lesser degrees. It would be interesting to present the agreement on detectable or undetectable nodules of these three techniques and to compare a particular profile of nodules showing disagreement among these three approaches within undetectable nodules. Those attempts would make it easier for radiologists to grasp the pros and cons when applying these three techniques at lung cancer screening with LDCT and to have greater confidence when choosing DLIR for screening.As a more basic question, the detectability performance for the subsolid nodule type was only 48.2% even in DLIR, with even lower performance when combined with other factors such as participant body mass index higher than 25 kg/m2 or nodule size less than 10 mm. Also surprisingly, 5.4% of nodules of at least 10 mm were undetectable. Ultimately, these issues still remain challenging, and additional technical advances are needed.What are the implications of the article’s findings for further clinical practice? The study by Jiang et al (10) adds to the growing body of clinical research now positioning the use of DLIR in a variety of lung disease situations—especially those that require repeated use of ULD CT, like COVID-19.Disclosures of conflicts of interest: H.Y.L. Activities related to the present article: disclosed no relevant relationships. Activities not related to the present article: is a consultant for Roche; has grants with Johnson & Johnson and Lunit; has stock options in Coreline Soft. Other relationships: disclosed no relevant relationships.References1. National Lung Screening Trial Research Team; Aberle DR, Adams AM, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med 2011;365(5):395–409. Crossref, Medline, Google Scholar2. Kim HJ, Park SY, Lee HY, Lee KS, Shin KE, Moon JW. Ultra-low-dose chest CT in patients with neutropenic fever and hematologic malignancy: image quality and its diagnostic performance. Cancer Res Treat 2014;46(4):393–402. Crossref, Medline, Google Scholar3. Udayasankar UK, Li J, Baumgarten DA, Small WC, Kalra MK. Acute abdominal pain: value of non-contrast enhanced ultra-low-dose multi-detector row CT as a substitute for abdominal radiographs. Emerg Radiol 2009;16(1):61–70. Crossref, Medline, Google Scholar4. Padole A, Ali Khawaja RD, Kalra MK, Singh S. CT radiation dose and iterative reconstruction techniques. AJR Am J Roentgenol 2015;204(4):W384–W392. Crossref, Medline, Google Scholar5. Willemink MJ, Noël PB. The evolution of image reconstruction for CT-from filtered back projection to artificial intelligence. Eur Radiol 2019;29(5):2185–2195. Crossref, Medline, Google Scholar6. Geyer LL, Schoepf UJ, Meinel FG, et al. State of the art: iterative CT reconstruction techniques. Radiology 2015;276(2):339–357. Link, Google Scholar7. Verdun FR, Racine D, Ott JG, et al. Image quality in CT: from physical measurements to model observers. Phys Med 2015;31(8):823–843. Crossref, Medline, Google Scholar8. Greffier J, Frandon J, Larbi A, Beregi JP, Pereira F. CT iterative reconstruction algorithms: a task-based image quality assessment. Eur Radiol 2020;30(1):487–500. Crossref, Medline, Google Scholar9. Park C, Choo KS, Jung Y, Jeong HS, Hwang JY, Yun MS. CT iterative vs deep learning reconstruction: comparison of noise and sharpness. Eur Radiol 2021;31(5):3156–3164. [Published correction appears in Eur Radiol 2021;31(6):4410-4411.] Crossref, Medline, Google Scholar10. Jiang B, Li N, Shi X, et al. Deep learning reconstruction shows better lung nodule detection for ultra–low-dose chest CT. Radiology 2022;303(1):202–212. Google ScholarArticle HistoryReceived: Dec 4 2021Revision requested: Dec 8 2021Revision received: Dec 8 2021Accepted: Dec 9 2021Published online: Jan 18 2022Published in print: Apr 2022 FiguresReferencesRelatedDetailsAccompanying This ArticleDeep Learning Reconstruction Shows Better Lung Nodule Detection for Ultra–Low-Dose Chest CTJan 18 2022RadiologyRecommended Articles Assessing Pulmonary Nodules by Using Lower Dose at CTRadiology2020Volume: 297Issue: 3pp. 708-709Deep Learning Reconstruction Shows Better Lung Nodule Detection for Ultra–Low-Dose Chest CTRadiology2022Volume: 303Issue: 1pp. 202-212Observer Performance for Detection of Pulmonary Nodules at Chest CT over a Large Range of Radiation Dose LevelsRadiology2020Volume: 297Issue: 3pp. 699-707Deep Learning–based Super-Resolution Algorithm: Potential in the Management of Subsolid NodulesRadiology2021Volume: 299Issue: 1pp. 220-221CT Noise-Reduction Methods for Lower-Dose Scanning: Strengths and Weaknesses of Iterative Reconstruction Algorithms and New TechniquesRadioGraphics2021Volume: 41Issue: 5pp. 1493-1508See More RSNA Education Exhibits Deep Learning Reconstruction In Chest CT - Issues Considered From The Dose Reduction And Image Quality In Clinical Cases -Digital Posters2021Deep Learning Based Image Reconstruction (DL-R) for CT: Can It Replace the Existing Image Reconstruction Techniques? Digital Posters2020Characteristics Of Deep Learning Reconstruction: Application In Clinical PracticeDigital Posters2021 RSNA Case Collection TI-RADS 3 nodule RSNA Case Collection2021Diffuse idiopathic pulmonary neuroendocrine cell hyperplasiaRSNA Case Collection2020Acute Pulmonary EmbolusRSNA Case Collection2021 Vol. 303, No. 1 Metrics Altmetric Score PDF download
Referência(s)