Lung Cancer

Introduction

There were 1.8 million new instances of lung cancer and 1.6 million deaths associated with the disease in 2012.Because of this, lung cancer is the second most prevalent cause of cancer-related deaths in males and the most common cause of cancer-related deaths overall.mortality among women from cancer following a diagnosis of breast cancer. The bulk of lung cancer cases (85%) are caused by long-term smoking, with only 10-15% of instances occurring in individuals who have never smoked.These cases were frequently brought on by a confluence of hereditary factors and exposure to air pollution, asbestos, radon gas, secondhand smoke, or other sources.Lung cancer ranks 15th among Saudi women and 5th among Saudi men in terms of frequency of occurrence in Saudi Arabia. In 2013, data indicated.

Methods

Training of the LCP-CNN The NLST dataset was used to train the LCP-CNN. The NLST’s eligibility criteria and study design have already been explained.^[7] This dataset had 10,368 people in total: There were 1058 people with lung cancer and 9310 participants with merely benign lung nodules. We classified as malignant any nodules in the training set that could be conclusively connected in hindsight to a diagnosis of lung cancer (N = 932 in 575 patients). All nodules in patients (N = 14,761 in 5972 participants) who were not diagnosed with lung cancer at the time of the NLST (screening and follow-up till seven years after baseline) were included as benign nodules.

The training set’s specifications are available in print ^[7]. Using this data, the LCP- CNN was trained to distinguish between benign and malignant nodules straight from the CT image by the use of supervised learning methodology. The algorithm adjusts its parameters until its predictions match the actual diagnosis of the nodules, which is how it learns. After being trained, the model (i.e., the LCP-CNN classifier generates a malignancy score per nodule) can forecast the likelihood that a new node would be benign or cancerous. Next, using a goal sensitivity of 100% (i.e., no malignancies missed), we determined a threshold based on the malignancy score on the NLST data to construct a benign nodule rule-out test.

An eight-fold cross-validation method was used to define the rule-out thresholds. Validation of the LCP-CNN Using lung nodules that were unintentionally found on thoracic CT images, we validated the LCP-CNN software on patients from the Early Lung Cancer Diagnosis Using Artificial Intelligence and Big Data (LUCINDA) project. Three hospitals were used to recruit patients: University Medical Centre Groningen in Groningen, The Netherlands (site A); Heidelberg University Hospital in Heidelberg, Germany (including Thoraxklinik Heidelberg), a tertiary referral centre for patients with lung cancer; and Oxford University Hospitals in Oxford, United Kingdom (including Royal Berkshire Hospital) (site C). The Appendix contains descriptions of the inclusion and exclusion criteria. Every site received permission from the local ethics committee.

A thoracic CT scan was used to choose data retroactively depending on the presence of pulmonary nodules. Every nodule had a diameter of five to fifteen millimetres. Clinicians, who had access to the ground truth diagnosis determined by the trial protocol, identified, contoured, and classified as benign or malignant each reported lung nodule and lung carcinoma (see Appendix). The CT data contained a range of scanner manufacturers, clinical indications, and diverse scan parameters. Thoracic CT scans with normal dosage, low dose, contrast-enhanced, and non-contrast enhanced were all included. The Appendix contains a description of CT acquisitions. Statistics Using the NLST data as training material, the LCP-CNN produced a malignancy score for every nodule in the external validation dataset.

Area-Under-the-ROC-Curve analysis (AUC) was used to assess overall performance. The percentage of benign nodules that the software properly classified as benign—known as benign rule-out performance—was computed by measuring the sensitivity and specificity at the pre-established score threshold.

Results

Nodule characteristics The validation set contained a total of 2106 unique nodules (205 malignant, 9.7%) from 1650 unique individuals (201 lung cancer patients, 12.2%) (Fig. 1). 489 patients (29.6%) were female, and the median patient age was 63.0 (range: 19–94). Table 1 lists the characteristics of nodules. See Tables A1 and A2 in the Appendix for more precise data on nodule location and size. Performance per site The LCP-CNN’s performance in identifying benign nodules varied depending on the site. Between the three centres, the overall AUC (Fig. 2) was 94.5 % (95%CI 92.6–96.1). The AUCs were 98.5 % (95 %CI 96.6–99.6), 88.1 % (95 %CI 84.5–91.3), and 97.7 % (95 %CI 95.9–99.0) for sites A, B, and C, respectively.

Centres A and C performed significantly better than centre B when comparing the AUCs between the centres (p < 0.01 and Fig. 1). *Flowchart* Technical exclusion occurred when there were significant motion artefacts surrounding the nodule, missing slices, or the CT scan was not the initial, primary, or axial CT scan. Benign-rule out performance With examples found in Figure A1 Appendix, the pre-defined score threshold for a benign nodule yielded an overall sensitivity of 99.0 % (95 %CI: 97.5 %–100.0 %) and specificity of 22.1% (95 %CI: 20.2 %–24.9 %). 99.5 percent had a negative predictive value. By applying the benignity threshold score, 420 benign nodules were accurately excluded (of which 11 were initially diagnosed based on histology, 5 on resolution at follow-up, 191 by expert opinion [i.e., perifissural nodules], 133 by volumetric stability after one year, and 80 by diameter stability after two years).

Two tumours were given false-negative results (Fig. A2 Appendix). In both cases, the carcinoids were round, smooth-surfaced, and weighed between 7 and 8 mm. When people were taken into account rather than only nodules, the software properly classified 18.5% (95 %CI: 16.5%–20.6%) of the patients as benign. where a patient’s highest-scoring nodule was considered when making a choice. The central lobe contained the ruledout nodules, which were frequently smaller than 8.0 mm (Table A2 Appendix).

Discussion

After being trained on lung nodule participants from the NLST dataset, our LCP-CNN performed exceptionally well in identifying benign nodules, accurately ruling out malignancy in one-fifth of patients with small-to-intermediate sized nodules. This suggests the potential utility of CNNs for incidentally discovered lung nodules in lung cancer risk prediction decision support. With a very high sensitivity of 99.0% in our study, CT scans can be ruled out, saving many patients from needless workups that include imaging and invasive treatments. It needs to be prospectively validated in a lung cancer screening programbeforeitcanbethoughttobeappliedhereaswell.Prior research on AI has demonstrated promise and concentrated on optimising the percentage of malignancies accurately characterised (high positive-predictive value).

^[8] The majority of these studies included big nodules up to 30 mm in diameter instead of the clinically relevant small-to-intermediate sized nodules, which limits the clinical usefulness of these tools’ sensitivity, but it is still moderate. The method we have verified in this work is not to use AI to identify lung tumours, but rather to use our AI system, the LCP-CNN, to identify benign nodules with high certainty and to recommend that these nodules may be ruled out, hence avoiding the need for more workup. By applying a cutoff point that is specified separately from the NLST data, We have demonstrated that approximately 18.5% of patients would not need additional work-up for this nodule if 22.1% of incidentally discovered benign nodules could be appropriately identified.

It has previously been demonstrated that the LCP-CNN performs better than the Brock University nodule classification model and can stratify nodules into high- and low malignancy risk categories when validated in separate datasets ^{[6, 7]}. While 22% of benign nodules were ruled out by the LCP-CNN, two cancers were overlooked. Both were typical tiny carcinoids, with a diameter of 7 and 8 mm, respectively. These are low-grade neuroendocrine lung tumours, accounting for 1-2 percent of all lung cancers. ^[9] The smooth surfaces and spherical shape of both carcinoids could be the reason for the false-negative result. outcome of the LCP-CNN. The three centres utilised identical model risk prediction score criteria to achieve a 97% target sensitivity; nevertheless, the LCP-CNN’s performance differed marginally amongst the patients in each centre.

The utilisation of distinct scan parameters and the fluctuating lung cancer incidence among patients, which ranges from 2.9% to 24.0% among the three centres, could perhaps account for this discrepancy. The majority of patients in centres A and C had lung malignancies and incidentally discovered nodules, while centre B is a tertiary referral centre that looks into more challenging cases with a high pre-test chance of malignancy. The centres with the biggest difference in median size between benign nodules and lung malignancies and the lowest risk of lung cancer were found to have performed well. However, Considering that 25% of the cases in the external dataset were from a tertiary referral centre, where patients primarily had nodules highly suggestive of being malignant, the software’s overall performance was quite good.

Standardised scanning parameters and quality assurance are required in a lung cancer screening program ^[10], and our LCP-CNN might do even better. In conclusion, the LCP-CNN performed exceptionally well in identifying benign lung nodules in a multi-center external dataset, ruling out malignancy with high sensitivity in roughly one-fifth of patients with intermediate-sized nodules. The LCP-CNN was trained on participants with lung nodules from the NLST dataset. Declaration of Competing Interest None. Acknowledgements This activity has received funding from the European Institute of Innovation and Technology (EIT) under grant agreement No 17189. This European body receives support from the Horizon 2020 research and innovation programme.

References

1. N. Horeweg, J. van Rosmalen, M.A. Heuvelmans,C.M. van der Aalst, R. Vliegenthart, E.Th Scholten, K. Ten Haaf, K. Nackaerts, J.W. Lammers, C. Weenink, H.J. Groen, P. van Ooijen, et al., Lung cancer probability in patients with CT-detected pulmonary nodules: a prespecified analysis of data from the NELSON trial of low-dose CT screening, Lancet Oncol. 15 (2014) 1332–1341, https:// doi.org/S1470-2045(14)70389-4 [pii].
2. H. MacMahon, D.P. Naidich, J.M. Goo, K.S. Lee,A.N.C. Leung, J.R. Mayo, A. C. Mehta, Y. Ohno, C.A. Powell, M. Prokop, G.D. Rubin, C.M. Schaefer-Prokop, W. D. Travis, P.E. Van Schil, A.A. Bankier, Guidelines for management ofincidentalpulmonarynodulesdetectedonCTimages: from the Fleischner Society 2017, Radiology 161659 (2017), https://doi.org/10.1148/radiol.2017161659
3. American College of Radiology, Lung CT Screening Reporting and Data System (Lung-RADS), July 7, 20, 2014, http://www.acr.org/Quality-Safety/Resources / LungRADS.
4. F. Ciompi, K. Chung, S.J. Van Riel, A.A.A. Setio, P.K. Gerke, C. Jacobs, E. Th Scholten, C. Schaefer-Prokop, M.M.W. Wille, A. Marchian`o, U. Pastorino, M. Prokop, B. Van Ginneken, Towards automatic pulmonary nodule management in lung cancer screening with deep learning, Sci. Rep. 7 (2017), https://doi.org/ 10.1038/ srep46479.
5. H. Peschl, D. Han, P. Van Ooijen, M. Oudkerk, et al., Lung Cancer prediction using deep learning software: validation on independent multi-centre data, J. Thorac. Oncol. 13 (2018) S428, https://doi.org/10.1016/j jtho.2018.08.489.
6. D.R. Baldwin, J. Gustafson, L. Pickup, C. Arteta, P. Novotny, J. Declerck, T. Kadir, C. Figueiras, A. Sterba, A. Exell, V. Potesil, P. Holland, H. Spence, A. Clubley, E. O’Dowd, M. Clark, V. Ashford-Turner, M.E.J. Callister, F.V. Gleeson, External validation of a convolutional neural network artificial intelligence tool to predict malignancy in pulmonary nodules, Thorax (2020), https://doi org/10.1136/ thoraxjnl-2019-214104.
7. P.P. Massion, S. Antic, S. Ather, C. Arteta, J. Brabec, H. Chen, J. Declerck, D. Dufek, W. Hickes, T. Kadir, J. Kunst, B.A. Landman, R.F. Munden, P. Novotny, H. Peschl, L.C. Pickup, C. Santos, G.T. Smith, A. Talwar, F. Gleeson, Assessing the accuracy of a deep learning method to risk stratify indeterminate pulmonary nodules, Am. J. Respir. Crit. Care Med. (2020), https://doi.org/10.1164/ rccm.201903-0505oc.
8. D. Ardila, A.P. Kiraly, S. Bharadwaj, B. Choi, J.J. Reicher, L. Peng, D. Tse, M. Etemadi, W. Ye, G. Corrado, D.P. Naidich, S. Shetty, End-to-end lung cancer screening with threedimensional deep learning on low-dose chest computed tomography, Nat. Med. 25 (2019) 954–961, https://doi org/10.1038/s41591-019- 0447-x.
9. B. Melosky, Advanced typical and atypical carcinoid tumours of the lung: management recommendations, Curr. Oncol. 25 (2018) S86–S93, https://doi.org/ 10.3747/ co.25.3808.
10. M. Oudkerk, A. Devaraj, R. Vliegenthart, T. Henzler,H. Prosch, C.P. Heussel, G. Bastarrika, N. Sverzellati, M. Mascalchi, S. Delorme, D.R. Baldwin, M. E. Callister, N. Becker, M.A. Heuvelmans, W. Rzyman, M.V. Infante, U. Pastorino, J. H. Pedersen, E. Paci, S.W. Duffy, H. de Koning, J.K. Field, European position statement on lung cancer screening, Lancet Oncol. 18 (2017) e754–e766, https:// doi.org/10.1016/S1470-2045(17)30861-6. Research Article

This is a text version generated from the article. For the formatted version of record (with original tables & figures), download the PDF →