Start Your Search
MA10 - Emerging Technologies for Lung Cancer Detection (ID 129)
- Event: WCLC 2019
- Type: Mini Oral Session
- Track: Screening and Early Detection
- Presentations: 2
- Now Available
MA10.06 - Randomized Clinical Trial with Computer Assisted Diagnosis (CAD) Versus Radiologist as First Reader of Lung Screening LDCT (Now Available) (ID 2102)
15:15 - 16:45 | Presenting Author(s): Ren Yuan
CAD has been studied extensively in lung nodule detection while its value in lung cancer screening has not been tested in a prospective randomized clinical study. We aim to evaluate the value of a CAD in radiologist work-flow and for quality assurance in reading and reporting lung cancer screening LDCT.
Between August 2016 and February 2019, 1386 ever smokers were enrolled in the BC Lung Screen Trial. The median follow-up was 10 months. Their chest CTs were randomized to CAD reading first arm using a CAD system (Philips IntelliSpace Portal) (n=741), or Radiologist reading first (RAD) arm (n=645). In CAD-1st arm, a radiologist read CTs with the CAD findings displayed concurrently, accepting, rejecting and adding nodule(s). Radiologist’s reading time was recorded, and management recommendation was made using the automatically generated PanCan lung nodule risk score (N Engl J Med 2013; 369:910-919). In Rad-1st arm, the radiologist read the CT without using CAD, gave the management recommendation using Lung-RADS, and the reading time was recorded. Then the radiologist turned on CAD annotations to accept, reject and add nodule(s). The PanCan nodule risk scores were generated. Nodule management was categorized into 3 groups: I: Scheduled follow-up CT ≥1yr for those with no or very low risk lung nodules; II: Early recall CT <1 yr; or III: Referral to clinical diagnostic pathway for suspicious malignancy.
Radiologist’s reading time was shorter in CAD-1st than Radiologist-1st arm (9±3 vs. 10±3 minutes, p<0.01). The time saved was greatest for Group I scans (85% of workload) (8±3 vs. 10±3 minutes, p<0.01). In 20/741 (2.7%) participants in CAD-1st arm, the additional nodule added by the radiologist upgraded the patient’s management; 5 of 20 were later confirmed to be malignant. Two of 5 were >3cm masses, the other three included a 19 mm GGO and two solid ones abutting vessels. In 1/645 (0.15%) participants in Radiologist-1st arm, the additional nodule detected by CAD upgraded the patient’s management from Group I to II. Over 31-months of follow-up, 29 cancers (2.1%) have been detected, and 1 of 29 (3.4%), a 5 mm solid nodule in the left lower lobe abutting the fissure and vessels, was missed by both radiologist and CAD.
CAD saves radiologist’s time in reading large numbers of screening LDCT especially in those with no or very low risk lung nodules. However, reading by experienced radiologist is still needed.
Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.
MA10.09 - Evaluation of the Clinical Utility of the PanCan, EU-NELSON and Lung-RADS Protocols for Management of Screen Detected Lung Nodules at Baseline (Now Available) (ID 2137)
15:15 - 16:45 | Author(s): Ren Yuan
Several protocols are available to guide management of lung nodules identified by the first (baseline) low-dose screening CT. It is important to objectively assess their clinical utility, health care resource utilization and potential harms. We aim to compare the PanCan (NEJM 2013;369:908 & J Thorac Oncol 2018; 13(10): S362-S363), EU-NELSON (Lancet Oncol. 2017 Dec;18(12):e754-e766 & Lung Cancer 2006;54:177-184) and Lung-RADS(https://www.acr.org/Clinical-Resources/Reporting-and-Data-Systems/Lung-Rads) lung nodule management protocols on our data set from two sites of the International Lung Screen Trial (ILST), in Vancouver, Canada and Perth, Western Australia.Method
Ever smokers age 55 to 80 years were enrolled into ILST if they had a ≥30 pack-years smoking history and smoked within 15 years or if their PLCO m2012 6 year lung cancer risk was ≥1.51%. The participants were managed via the PanCan lung nodule risk based protocol. The NELSON and Lung-RADS nodule protocols were applied to the ILST data set. The potential difference in the proportion of the participants having an early recall CT scan (< 1 year) or referral to a clinical diagnostic pathway was compared between the PanCan, NELSON, Lung-RADS protocols. The participants were divided into 3 groups: Group 1 (next scheduled annual/biennial CT) included PanCan CAT 1, 2, NELSON NODCAT I, II, Lung-RADS CAT 1, 2. Group 2 (early recall CT <1 year) included PanCan CAT 3, NELSON NODCAT III, Lung-RADS CAT 3, 4A and Group 3 (Diagnostic Pathway) included PanCan CAT 4, 5, NELSON NODCAT IV (solid nodule), Lung-RADS CAT 4B, 4X. The number of participants and the lung cancer rate in each group was compared between the three protocols.Result
A total of 1386 participants with a median follow-up of 10 months (ranging from 4-31 mos) were evaluated. The results are shown in Table 1.
PanCan selected the fewest individuals to early recall (Group 2 & 3) versus NELSON p=0.004 and detected the same number of lung cancers as did the NELSON and more than by Lung-RADS.
In addition, 81% of the PanCan group 1 participants were triaged to biennial repeat screening instead of annual screening in the NELSON and Lung-RADS protocols, which has substantial resource utilization and radiation exposure implications.
The personalized risk-based PanCan Protocol may decrease resource utilization and potentially minimize risk of screening for participants.
OA06 - Refining Lung Cancer Screening (ID 131)
- Event: WCLC 2019
- Type: Oral Session
- Track: Screening and Early Detection
- Presentations: 1
- Now Available
- Moderators:Tomasz Grodzki, Lluis Esteban Tejero
- Coordinates: 9/09/2019, 11:00 - 12:30, Hilton Head (1978)
OA06.01 - Comparison Between Radiomics-Based Machine Learning and Deep Learning Image Classification for Sub-Cm Lung Nodules (Now Available) (ID 2810)
11:00 - 12:30 | Author(s): Ren Yuan
New clinical challenges have arisen from the recent recognition for an improved mortality of cancers via lung cancer screening using LDCT. A particular challenge for physicians and CADx systems is the classification and prediction of behavior for sub-cm lung nodules that are frequently present in screening CT scans. By predicting and classifying the behavior of these small nodules, we can identify potential cancerous nodules into the earlier stages of malignancy making them more easily treatable.Method
We have evaluated a multitude of image processing techniques to assist in CADx systems for these small nodules such as Radiomic feature-based machine learning algorithms (linear discriminant analysis) as well as leveraging pretrained convolution neural networks such as VGG19 and InceptionV3 using deep learning/transfer learning techniques. The linear discriminate Radiomic analysis (LDA) classified a sample of CT imaged nodules (n=514) using quasi-volumetric nodule data (images of the nodules from CT slices above and below the central slice) into three discriminate categories: cancerous (clinically confirmed, n = 140) versus resolved (not present in follow up CT scans, n=107) versus stable (a negligible change in shape, texture, size in multi year follow up CT scans, n=267). Each nodule was segmented from the original CT scan using an inhouse lung CT image segmenation routine. This routine generated 2167 discrete CT nodule images upon which 133 Radiomic shape and texture features were calculated.Result
The LDA Radiomic analysis correctly classified the individual nodual sections with an accuracy of 75.1% (jackknife - leave one out result) using only 18 features predefined traditional image analysis features (4 shape feature(s), 14 texture feature(s)) for cancer vs resolved + stable nodules. Requiring that more than or equal to 50% of sections from a nodule be classified as cancer for the nodule to be classified as cancer individual nodules could be correctly classified with an 80% accuracy.
The leveraged pretrained networks (VGG19, and InceptionV3) trained using standard data augmentation and finetuning techniques, trained on this same quasi-volumetric image data as a binary classification task (malignant vs. benign nodules) achieved an average classification accuracy of 71% and 75% respectively through 10-crossfold validation.Conclusion
Machine learning using 18 Radiomics features was able to classify 75.1% of the 2167 CT nodule images (up to 5 images/CT slices per nodule) and 80% of the nodules correctly. The best of the Deep Learning networks achieved almost equivalent results.
The image classification deep neural network results suggest the implementation of more advanced regularization and initialization deep learning techniques to further refine the decision boundaries for these pretrained networks might be benefitial. We believe the development of visualization neural network software to highlight the defining nodule features during classification would clinically assist in providing context clues for nodule diagnosis.
This work has been supported by TFRI project ref:1068