Lung Cancer

Cancer is a leading cause of death worldwide and accounted for 8.8 million deaths (around 16% of all deaths) in 2015. Lung cancer is the most common cause of cancer-related death in men and women and accounted for 1.7 million deaths in 2015 and accounts for more annual deaths than breast, colon and prostate cancers combined. A major risk factor for lung cancer is tobacco use, but also alcohol consumption, air pollution, poor diet and physical inactivity are main risk factors. Prevention of lung cancer is difficult since symptoms usually arise when the cancer is already in a very late stage. 5-Year survival rates for stage IV lung cancers are reported to be as low as 10%. However, prognosis is a lot better when lung cancer is detected in an early stage. 5-year survival rates for stage I lung cancers are reported to be between 60-80%.

Patient559 metSigaretten cropped.png
Fig. 1. An axial slice of a thin-slice low-dose CT acquisition. The pack of sigarettes which is visible in this examination nicely illustrates the main risk factor for lung cancer: tobacco use.

Low-dose CT Lung Cancer Screening

Because the prognosis is so much better when lung cancer is detected in an early stage, efforts are under way to start low-dose CT screening of high-risk subjects to detect lung cancers in an early stage. Randomized clinical trials (RCTs) are needed to assess the efficacy of lung cancer screening with low-dose chest CT in reducing lung cancer mortality. As of 2000, different lung cancer screening trials using low-dose CT have been conducted. Table 1 shows an overview of the three largest randomized lung cancer trials in the world including the number of subjects, inclusion criteria, control group characteristics and possible result.

Trial # Participants Inclusion criteria Control arm Start Date Outcome
NELSON 15,523 16 cigarettes a day , >= 26 years or >= 11 cigarettes a day >= 31 years

> 50 years old

No screening August 2003 To be published
NLST 53,454 > 30 pack years, >55 years old Chest X-Ray (CXR) September 2002 20% lung cancer mortality reduction in CT arm
DLCST 4,104 >50 yrs old, >20 pack years No screening October 2004 No statistically significant effect on lung cancer mortality found, but post hoc

high-risk subgroup analyses showed consistency with NLST results

ITALUNG 3,206 55-69 years old, smoker or ex-smoker (at least 20 pack-years in the last 10 years) No screening September 2003 30% reduction in lung cancer specific mortality (not significant)

While low-dose CT screening is expected to reduce lung cancer mortality, this was not found in lung cancer screening trials until November 2010. On the 15th of November, the National Lung cancer Screening Trial (NLST) in the United States reported a 20% mortality reduction in high-risk subjects. In this study, the study group received an annual low-dose CT scan for three years while the control group only received a chest radiograph (CXR), also for three years. This again raised the attention to low-dose CT screening for high-risk subjects. But there are still several issues which need to be solved before screening for lung cancer can be implemented. Among the issues are cost-effectiveness, selection of high-risk subjects, frequency of screening and overdiagnosis. Overdiagnosis is the diagnosis of disease that will never cause symptoms or death during a patient's lifetime. New randomized control trials are planned to investigate these aspects.

Pulmonary nodules

Early stage lung cancers are represented by focal pulmonary lesions between 3 and 30 mm in diameter which show generally a high contrast with the surrounding lung parenchyma on a CT scan. An example of this can be seen in Fig. 2.

Fig. 2. An axial slice of a thin-slice CT acquisition (low-dose). The yellow arrow indicates a solid pulmonary nodule.

Three different categories of pulmonary nodules are defined: solid nodules, part-solid nodules and non-solid nodules (also called ground glass nodules). The last two categories can also be combined into the category subsolid nodules. The prevalence and malignancy rates of these different types of nodules differ. In the Early Lung Cancer Action Project (ELCAP) study, at baseline, the prevalence of solid nodules, part-solid nodules and non-solid nodules was 81%, 7% and 12%, respectively. However, the malignancy rates for the three different nodule types were 7%, 62% and 18% respectively. This shows that subsolid nodules have a much higher malignancy rate than solid nodules. As a consequence, 51% of the cancers found at baseline in the ELCAP study actually originated from subsolid nodules.

Nodule detection in Chest CT

In order to diagnose lung cancer in an early stage, it is important to detect pulmonary nodules in chest CT scans and assess the risk of malignancy. In order to detect also small pulmonary nodules around 3-5 mm in diameter, it is important that thin-slice CT examinations (around 1mm) are acquired. These thin-slice acquisitions generates between 200 and 400 axial slices per scan which have to be inspected by a thoracic radiologist for pulmonary nodules. As shown in previous studies, this process is time-consuming and there is a risk of missed nodules. Therefore, computer-aided detection (CAD) software is being developed to aid the radiologist. The purpose of these algorithms is to automatically detect the pulmonary nodules in CT scans and point the radiologist to these suspicious areas. This has received great interest in the medical image analysis field and many nodule CAD papers have been published. At present, although CAD systems already reach a good sensitivity, they are still not as good as an experienced radiologist and generate too many false positives.


In this project, we focus on early detection of lung cancer. Much of our work is about extending and improving our current nodule CAD algorithms and automating lung cancer screening, but we also do clinical research. To help us achieve our goals, we have developed a high-throughput workstation for lung cancer screening (see more info below), which incorporates many of the algorithms we develop. This project is done in close collaboration with MeVis Medical Solutions AG and has led to the release of Veolity, a optimized workstation solution for lung cancer screening. Below, you can find information about the challenges we organized and our workstation.

ANODE09 challenge

In 2009, Bram van Ginneken organized the Automatic NOdule DEtection (ANODE09) study. ANODE09 is an initiative to compare commercial and non-commercial systems that perform automatic detection of pulmonary nodules in chest CT scans on a single common database, with a single evaluation protocol. Data were provided by the NELSON study, the largest CT lung cancer screening trial in Europe. The study consisted of a database of 55 scans and a web-based framework for objective evaluation of nodule detection algorithms. Any team can upload results to facilitate benchmarking. The performance of six algorithms are compared; five from academic groups and one commercially available system. A method to combine the output of multiple systems is proposed and the results showed a substantial performance difference between algorithms, and demonstrate that combining the output of algorithms leads to marked performance improvements. In Fig. 3, you can see the FROC curves of the different systems and the combination of systems.

Fig. 3. FROC curves of all six systems and three combinations. The horizontal axis is logarithmic and covers four orders of magnitude.

A special session devoted to the ANODE09 study was held at the 2009 CAD Conference of SPIE Medical Imaging and a joint paper with the first results of the study has been accepted by Medical Image Analysis.

LUNA16 challenge

In 2016, Arnaud Setio, Alberto Traverso, Bram van Ginneken, and Colin Jacobs organized the LUNA16 challenge. LUNA16 used 888 CT scans from the Lung Image Database Consortium image collection (LIDC-IDRI) data set, the largest publicly availabe reference data set for nodule CAD. Again, computer vision researchers from around the world were invited to participate in this challenge. All top performing systems used deep learning. In Fig. 4, you can see the FROC curves of the top 7 systems in the NDET track of LUNA16.

Fig. 3. FROC curves of all systems in the NDET track of the LUNA16 challenge.

A special session devoted to the LUNA16 study was held at the 2016 ISBI Conference in Prague, Czech Republic and a joint paper with the first results of the study has been published in Medical Image Analysis.

Data Science Bowl 2017

In 2017, the third Data Science Bowl, a yearly competition organized by Kaggle, was about detecting lung cancer from CT scans. A total of one million dollar was made available as prize money. Bram van Ginneken, Colin Jacobs and Arnaud Setio coorganized this challenge, which received a lot of attention, see for example here. The results of the competition can be found here

Workstation for lung screening

We believe that in its current form, large scale introduction of CT lung screening would put an enormous burden on radiologists. Building upon our clinical and technical experience in reading, image analysis and data processing for large screening trials in Europe (over 30,000 CT scans from 10,000 participants) and a careful review of the existing commercially available lung workstations, we are developing a new dedicated chest reading workstation with a number of innovations that allows for an optimized high throughput workflow to report on low dose chest CT scans. At present, the workstation is being used at several sites as a research prototype. A screenshot of the workstation can be seen in Fig. 4 and Fig. 5.

Workstation NoduleCAD cropped.png
Fig. 4. Nodule detection in workstation. The CAD system indicates a subtle ground glass nodule in the right lung.
GGNoverTime cropped.png
Fig. 5. Nodule growth assessment of a ground glass nodule over time in workstation.


This project is funded and executed in close collaboration with Fraunhofer MEVIS and MeVis Medical Solutions AG, Bremen, Germany.



Key publications

  • A.A.A. Setio, A. Traverso, T. de Bel, M.S.N. Berens, C. v. d. Bogaard, P. Cerello, H. Chen, Q. Dou, M.E. Fantacci, B. Geurts, R. v. d. Gugten, P.A. Heng, B. Jansen, M.M.J. de Kaste, V. Kotov, J.Y.-H. Lin, J.T.M.C. Manders, A. Sóñora-Mengana, J.C. García-Naranjo, E. Papavasileiou, M. Prokop, M. Saletta, C.M. Schaefer-Prokop, E.T. Scholten, L. Scholten, M.M. Snoeren, E.L. Torres, J. Vandemeulebroucke, N. Walasek, G.C.A. Zuidhof, B. v. Ginneken and C. Jacobs. "Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography images: The LUNA16 challenge", Medical Image Analysis 2017;42:1-13. Abstract/PDF DOI arXiv PMID

  • K. Chung, C. Jacobs, E.T. Scholten, J.M. Goo, H. Prosch, N. Sverzellati, F. Ciompi, O.M. Mets, P.K. Gerke, M. Prokop, B. van Ginneken and C.M. Schaefer-Prokop. "Lung-RADS Category 4X: Does It Improve Prediction of Malignancy in Subsolid Nodules?", Radiology 2017;284(1):264-271. Abstract/PDF DOI PMID

  • A.J. Ritchie, C. Sanghera, C. Jacobs, W. Zhang, J. Mayo, H. Schmidt, M. Gingras, S. Pasian, L. Stewart, S. Tsai, D. Manos, J.M. Seely, P. Burrowes, R. Bhatia, S. Atkar-Khattra, B. van Ginneken, M. Tammemagi, M.S. Tsao, S. Lam and the Pan-Canadian Early Detection of Lung Cancer Study Group. "Computer Vision Tool and Technician as First Reader of Lung Cancer Screening CT Scans", Journal of Thoracic Oncology 2016;11:709-717. Abstract/PDF DOI PMID

  • A.A.A. Setio, F. Ciompi, G. Litjens, P. Gerke, C. Jacobs, S. van Riel, M.W. Wille, M. Naqibullah, C. Sánchez and B. van Ginneken. "Pulmonary nodule detection in CT images: false positive reduction using multi-view convolutional networks", IEEE Transactions on Medical Imaging 2016;35(5):1160-1169. Abstract/PDF DOI PMID

  • S.J. van Riel, C.I. Sánchez, A.A. Bankier, D.P. Naidich, J. Verschakelen, E.T. Scholten, P.A. de Jong, C. Jacobs, E. van Rikxoort, L. Peters-Bax, M. Snoeren, M. Prokop, B. van Ginneken and C. Schaefer-Prokop. "Observer Variability for Classification of Pulmonary Nodules on Low-Dose CT Images and Its Effect on Nodule Management", Radiology 2015;277(3):863-871. Abstract/PDF DOI PMID

  • C. Jacobs, E.M. van Rikxoort, E.T. Scholten, P.A. de Jong, M. Prokop, C. Schaefer-Prokop and B. van Ginneken. "Solid, Part-Solid, or Non-solid?: Classification of Pulmonary Nodules in Low-Dose Chest Computed Tomography by a Computer-Aided Diagnosis System", Investigative Radiology 2015;50:168-173. Abstract/PDF DOI PMID

  • E.T. Scholten, P.A. de Jong, B. de Hoop, R. van Klaveren, S. van Amelsvoort-van de Vorst, M. Oudkerk, R. Vliegenthart, H.J. de Koning, C.M. van der Aalst, R.M. Vernhout, H.J.M. Groen, J.-W.J. Lammers, B. van Ginneken, C. Jacobs, W.P.T.M. Mali, N. Horeweg, C. Weenink, E. Thunnissen, M. Prokop and H.A. Gietema. "Towards a close computed tomography monitoring approach for screen detected subsolid pulmonary nodules?", European Respiratory Journal 2015;45:765-773. Abstract/PDF DOI PMID

  • C. Jacobs, E.M. van Rikxoort, T. Twellmann, E.T. Scholten, P.A. de Jong, J.M. Kuhnigk, M. Oudkerk, H.J. de Koning, M. Prokop, C. Schaefer-Prokop and B. van Ginneken. "Automatic Detection of Subsolid Pulmonary Nodules in Thoracic Computed Tomography Images", Medical Image Analysis 2014;18:374-384. Abstract/PDF DOI PMID

  • C. Jacobs, C.I. Sánchez, S.C. Saur, T. Twellmann, P.A. de Jong and B. van Ginneken. "Computer-Aided Detection of Ground Glass Nodules in Thoracic CT images using Shape, Intensity and Context Features", in: Medical Image Computing and Computer-Assisted Intervention, volume 6893 of Lecture Notes in Computer Science, 2011, pages 207-214. Abstract/PDF DOI PMID

  • B. van Ginneken, S.G. Armato, B. de Hoop, S. van de Vorst, T. Duindam, M. Niemeijer, K. Murphy, A.M.R. Schilham, A. Retico, M.E. Fantacci, N. Camarlinghi, F. Bagagli, I. Gori, T. Hara, H. Fujita, G. Gargano, R. Belloti, F.D. Carlo, R. Megna, S. Tangaro, L. Bolanos, P. Cerello, S.C. Cheran, E.L. Torres and M. Prokop. "Comparing and combining algorithms for computer-aided detection of pulmonary nodules in computed tomography scans: the ANODE09 study", Medical Image Analysis 2010;14:707-722. Abstract/PDF DOI PMID

  • K. Murphy, B. van Ginneken, A.M.R. Schilham, B.J. de Hoop, H.A. Gietema and M. Prokop. "A Large Scale Evaluation of Automatic Pulmonary Nodule Detection in Chest CT using Local Image Features and k-Nearest-Neighbour Classification", Medical Image Analysis 2009;13(5):757-770. Abstract/PDF DOI PMID