Evaluation of a novel deep learning-based classifier for perifissural nodules.

Published on Dec 2, 2020in European Radiology4.101
· DOI :10.1007/S00330-020-07509-X
Daiwei Han3
Estimated H-index: 3
(UMCG: University Medical Center Groningen),
Marjolein A Heuvelmans17
Estimated H-index: 17
(UMCG: University Medical Center Groningen)
+ 10 AuthorsRozemarijn Vliegenthart37
Estimated H-index: 37
(UMCG: University Medical Center Groningen)
To evaluate the performance of a novel convolutional neural network (CNN) for the classification of typical perifissural nodules (PFN). Chest CT data from two centers in the UK and The Netherlands (1668 unique nodules, 1260 individuals) were collected. Pulmonary nodules were classified into subtypes, including “typical PFNs” on-site, and were reviewed by a central clinician. The dataset was divided into a training/cross-validation set of 1557 nodules (1103 individuals) and a test set of 196 nodules (158 individuals). For the test set, three radiologically trained readers classified the nodules into three nodule categories: typical PFN, atypical PFN, and non-PFN. The consensus of the three readers was used as reference to evaluate the performance of the PFN-CNN. Typical PFNs were considered as positive results, and atypical PFNs and non-PFNs were grouped as negative results. PFN-CNN performance was evaluated using the ROC curve, confusion matrix, and Cohen’s kappa. Internal validation yielded a mean AUC of 91.9% (95% CI 90.6–92.9) with 78.7% sensitivity and 90.4% specificity. For the test set, the reader consensus rated 45/196 (23%) of nodules as typical PFN. The classifier-reader agreement (k = 0.62–0.75) was similar to the inter-reader agreement (k = 0.64–0.79). Area under the ROC curve was 95.8% (95% CI 93.3–98.4), with a sensitivity of 95.6% (95% CI 84.9–99.5), and specificity of 88.1% (95% CI 81.8–92.8). The PFN-CNN showed excellent performance in classifying typical PFNs. Its agreement with radiologically trained readers is within the range of inter-reader agreement. Thus, the CNN-based system has potential in clinical and screening settings to rule out perifissural nodules and increase reader efficiency. • Agreement between the PFN-CNN and radiologically trained readers is within the range of inter-reader agreement. • The CNN model for the classification of typical PFNs achieved an AUC of 95.8% (95% CI 93.3–98.4) with 95.6% (95% CI 84.9–99.5) sensitivity and 88.1% (95% CI 81.8–92.8) specificity compared to the consensus of three readers.
#1Pierre P. Massion (Vandy: Vanderbilt University)H-Index: 65
#2Sanja L. Antic (Vandy: Vanderbilt University)H-Index: 4
Last. Fergus V. Gleeson (University of Oxford)H-Index: 58
view all 20 authors...
Rationale: The management indeterminate pulmonary nodules (IPNs) remains challenging, resulting in invasive procedures and delays in diagnosis and treatment. Strategies to decrease the rate of unne...
15 CitationsSource
#1David R Baldwin (University of Nottingham)H-Index: 43
#2Jennifer Gustafson (Churchill Hospital)H-Index: 2
Last. Fergus V. Gleeson (Churchill Hospital)H-Index: 58
view all 19 authors...
Background Estimation of the risk of malignancy in pulmonary nodules detected by CT is central in clinical management. The use of artificial intelligence (AI) offers an opportunity to improve risk prediction. Here we compare the performance of an AI algorithm, the lung cancer prediction convolutional neural network (LCP-CNN), with that of the Brock University model, recommended in UK guidelines. Methods A dataset of incidentally detected pulmonary nodules measuring 5–15 mm was collected retrospe...
21 CitationsSource
#1Harry J. de Koning (EUR: Erasmus University Rotterdam)H-Index: 27
#2Carlijn M. van der Aalst (EUR: Erasmus University Rotterdam)H-Index: 21
Last. Matthijs Oudkerk (EUR: Erasmus University Rotterdam)H-Index: 94
view all 24 authors...
Abstract Background There are limited data from randomized trials regarding whether volume-based, low-dose computed tomographic (CT) screening can reduce lung-cancer mortality among male former and...
338 CitationsSource
#1Daiwei Han (UMCG: University Medical Center Groningen)H-Index: 3
#2Marjolein A Heuvelmans (UMCG: University Medical Center Groningen)H-Index: 17
Last. Matthijs Oudkerk (UG: University of Groningen)H-Index: 94
view all 12 authors...
Abstract Introduction In incidence lung cancer screening rounds, new pulmonary nodules are regular findings. They have a higher lung cancer probability than baseline nodules. Previous studies showed that baseline perifissural nodules (PFNs) represent benign lesions. Whether this is also the case for incident PFNs is unknown. This study evaluated newly detected nodules in the Dutch-Belgian randomized-controlled NELSON study with respect to incidence of fissure-attached nodules, their classificati...
3 CitationsSource
#1Ugo PastorinoH-Index: 27
#1Ugo PastorinoH-Index: 70
Last. Alfonso MarchianòH-Index: 31
view all 10 authors...
ABSTRACT Background The National Lung Screening Trial showed that lung cancer (LC) screening by three annual rounds of low-dose computed tomography (LDCT) reduces LC mortality. We evaluated the benefit of prolonged LDCT screening beyond 5years, and its impact on overall and LC specific mortality at 10years. Design The Multicentric Italian Lung Detection (MILD) trial prospectively randomized 4099 participants, to a screening arm (n=2376), with further randomization to annual (n=1190) or biennial ...
94 CitationsSource
#1Anton SchreuderH-Index: 3
#2Bram van GinnekenH-Index: 81
view all 9 authors...
The interreader agreement for classifying solid opacities of 5 to 10 mm in diameter as perifissural nodules was moderate at best. Misclassifications of malignancies as perifissural nodules occurred...
13 CitationsSource
#1Paul F. Pinsky (NIH: National Institutes of Health)H-Index: 61
#2Christina Bellinger (Wake Forest University)H-Index: 9
Last. David P. Miller (Wake Forest University)H-Index: 55
view all 3 authors...
ObjectivesLow-dose computed tomography lung cancer screening has been shown to reduce lung cancer mortality but has a high false-positive rate. The precision medicine approach to low-dose computed tomography screening assesses subjects’ benefits versus harms based on their personal lung cancer risk, where harms include false-positive screens and resultant invasive procedures. We assess the relationship between lung cancer risk and the rate of false-positive LDCT screens.MethodsThe National Lung ...
22 CitationsSource
#1Onno M. Mets (UU: Utrecht University)H-Index: 18
#2Kaman Chung (Radboud University Nijmegen Medical Centre)H-Index: 9
Last. Pim A. de Jong (UU: Utrecht University)H-Index: 56
view all 8 authors...
Objectives Perifissural nodules (PFNs) are a common finding on chest CT, and are thought to represent non-malignant lesions. However, data outside a lung cancer-screening setting are currently lacking.
17 CitationsSource
#1Marjolein A Heuvelmans (Medisch Spectrum Twente)H-Index: 17
#1Marjolein A Heuvelmans (Medisch Spectrum Twente)H-Index: 5
Last. Rozemarijn VliegenthartH-Index: 37
view all 12 authors...
Abstract Objectives To explore the relationship between nodule count and lung cancer probability in baseline low-dose CT lung cancer screening. Materials and Methods Included were participants from the NELSON trial with at least one baseline nodule (3392 participants [45% of screen-group], 7258 nodules). We determined nodule count per participant. Malignancy was confirmed by histology. Nodules not diagnosed as screen-detected or interval cancer until the end of the fourth screening round were re...
25 CitationsSource
#1Francesco Ciompi (Radboud University Nijmegen)H-Index: 23
#2Bartjan de Hoop (UU: Utrecht University)H-Index: 18
Last. Bram van Ginneken (Radboud University Nijmegen)H-Index: 81
view all 9 authors...
In this paper, we tackle the problem of automatic classification of pulmonary peri-fissural nodules (PFNs). The classification problem is formulated as a machine learning approach, where detected nodule candidates are classified as PFNs or non-PFNs. Supervised learning is used, where a classifier is trained to label the detected nodule. The classification of the nodule in 3D is formulated as an ensemble of classifiers trained to recognize PFNs based on 2D views of the nodule. In order to describ...
164 CitationsSource
Cited By0