University of Groningen
Pulmonary Nodules: 2D versus 3D evaluation in lung cancer screening
Han, Daiwei
DOI:
10.33612/diss.172563513
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2021
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Han, D. (2021). Pulmonary Nodules: 2D versus 3D evaluation in lung cancer screening. University of Groningen. https://doi.org/10.33612/diss.172563513
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han Processed on: 20-5-2021 Processed on: 20-5-2021 Processed on: 20-5-2021
Processed on: 20-5-2021 PDF page: 149PDF page: 149PDF page: 149PDF page: 149
CHAPTER
10
559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han Processed on: 20-5-2021 Processed on: 20-5-2021 Processed on: 20-5-2021
Processed on: 20-5-2021 PDF page: 150PDF page: 150PDF page: 150PDF page: 150
10
CHAPTER 10
152
Lung cancer is the most common cause of cancer-related mortality worldwide. Despite advances in treatment lung cancer continues to have high mortality rate. Early detection using computed tomography (CT) has been shown to improve the survival in lung cancer patients. The Dutch Belgian Randomized Lung Cancer Screening Trial (Dutch acronym: NELSON) is the first lung cancer screening trial, in which nodule management is based on nodule volume and volume-doubling time (VDT), in contrast to other trials that are based on nodule diameter assessment. The aim of this thesis is to compare volume-based with diameter-volume-based pulmonary nodule assessment methods, and evaluation and classification of perifissural nodules (PFN).
Part 1: 2D versus 3D assessment of pulmonary nodules in lung cancer screening
In chapter 2, evidence from the different European lung cancer screening trials were
reviewed and the recommendations from European medical societies on lung cancer screening were summarized. It was found that, although there is a great support for the implementation of lung cancer screening, there is still some variation regarding the recommendation from different European medical societies. Nevertheless, almost all medical societies prefer the use of volumetric assessment of pulmonary nodules over manual diameter assessment.
In lung cancer screening it is highly important to adequately differentiate malignant from benign nodules. Accurate evaluation of nodule growth during follow-up is one of the prerequisites for nodule distinction. In chapter 3, the literature of volumetric assessment of and diameter assessment of pulmonary nodules was reviewed. We have found accumulating evidence that semi-automatic volume measurements may be better suited to evaluate nodule malignancy due to its higher reproducibility, and sensitivity for nodule growth compared to diameter measurements. In chapter 4, we evaluated the influence of nodule margin on volume- and diameter-based measurement variability. We demonstrated for nodules of indeterminate size (50 mm3 – 500 mm3) that diameter-based assessment is particularly susceptible to the influence of nodule margin. The inter-reader variability for diameter-based assessment was especially poor for nodules with spiculated and irregular margins and exceeded the 1.5 mm growth cut-off from the U.S. lung cancer screening nodule management guideline Lung-RADS by 133% and 200%, respectively. This effect was much smaller for volume-based assessment. Poor measurement variability makes it difficult to accurately determine nodule growth(rate). Therefore, volume-based assessment should be preferred over diameter measurements for nodule size and growth determination in CT lung cancer screening.
As lung cancer screening could soon become much more widespread, the health risks caused by ionizing radiation from CT lung cancer screening have become a concern.
559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han Processed on: 20-5-2021 Processed on: 20-5-2021 Processed on: 20-5-2021
Processed on: 20-5-2021 PDF page: 151PDF page: 151PDF page: 151PDF page: 151
10
153
On the other hand, CT images produced using reduced radiation dose may hinder the performance of radiologists in nodule detection and evaluation. In chapter 5, we evaluated the influence of ultra-low dose CT (ULDCT) with tin filter and iterative reconstruction technique on lung nodule assessment. We found that at 83.1% reduced radiation dose, the detection sensitivity was nonsignificant between HRCT 88.8% (95%CI: 76.0-96.3) and ULDCT 95.5% (95%CI: 84.9-99.5). No significant difference in semi-automated volumetry and diameter measurement was found between HRCT and ULDCT. Thus, this study indicates that ULDCT with tin filter and iterative reconstruction may potentially be applicable in lung cancer screening.
Part 2: Classification of perifissural nodules
Perifissural nodules (PFNs) are a subcategory of solid nodules accounting for up to 30% of nodules in the baseline screening round of lung cancer screening studies. These nodules have defined characteristics and are thought to represent non-malignant nodules such as intrapulmonary lymph nodes. The NELSON trial has shown that new nodules detected in the incidence rounds have higher lung cancer probability compared to nodules detected at baseline. Whether new PFNs from incidence rounds can be safely excluded from further follow-up is unknown. In chapter 6, we evaluated new solid nodules from the NELSON study with respect to lung cancer probability, incidence, and classification of PFNs. We found that 4% of incident solid nodules were PFNS, none of the malignant nodules were classified as PFNs. This indicates that PFNs may safely be ruled out in incidence screening rounds. Furthermore, nodules that are fissure-attached but were not classified as PFN, had malignancy probability of 25%. These nodules will need short term follow-up and should be evaluated using semi-automatic volumetry for growth. In chapter 7, we reviewed previous studies on PFNs and their classification. We discussed variations of PFN definitions in the literature and concluded that agreement on an unambiguous definition for the classification of PFNs is needed for future studies on PFN.
Due to the characteristic appearance of typical PFNs, benign nature, and their prevalence, they are an interesting target for the training of a convolutional neural network (CNN) classifier for their rule out. In chapter 8, we evaluated the performance of a CNN for the classification of typical PFNs (PFN-CNN). We focused only on typical PFNs due to their less ambiguous definition and better inter-reader agreement by expert readers when compared to atypical PFNs. When compared to the consensus of three trained readers the CNN classifier achieved an AUC of 95.8 (95% CI 93.3-98.4), 95.6% (95% CI 84.9-99.5) sensitivity, and 88.1% (95% CI 81.8-92.8) specificity. The agreement between the PFN-CNN classifier and the readers is within the range of inter-reader agreement. This study indicates that CNN based classifiers have the potential to rule out benign nodules
559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han Processed on: 20-5-2021 Processed on: 20-5-2021 Processed on: 20-5-2021
Processed on: 20-5-2021 PDF page: 152PDF page: 152PDF page: 152PDF page: 152
10
CHAPTER 10
154
and improving reader efficiency. However, further evaluation of the PFN-CNN using an external dataset is required.
In chapter 9, the main findings of studies in this thesis are summarized. Clinical implications and future perspectives are discussed. Firstly, this thesis provides additional evidence supporting the use of volumetric assessment for the evaluation of pulmonary nodules in lung cancer screening. Semi-automatic volumetry is more accurate and reproducible than diameter assessment especially for non-smooth-margined nodules. Secondly, ULDCT has comparable nodule detection rate as HRCT, at almost a tenth of the radiation dose. Furthermore, there was no significant trade-off in nodule measurement and size classification in ULDCT when compared to standard dose CT. Thirdly, new fissure-attached PFNs detected in incidental screening rounds are benign and can be safely ruled-out from further follow-ups. Finally, a CNN-based nodule classifier seems to be promising in identifying benign PFNs for their rule-out.
559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han Processed on: 20-5-2021 Processed on: 20-5-2021 Processed on: 20-5-2021
Processed on: 20-5-2021 PDF page: 153PDF page: 153PDF page: 153PDF page: 153
10
559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han 559233-l-bw-Han Processed on: 20-5-2021 Processed on: 20-5-2021 Processed on: 20-5-2021