Deep Learning-Based Stage-Wise Risk Stratification for Early Lung Adenocarcinoma in CT Images: A Multi-Center Study
Abstract
:Simple Summary
Abstract
1. Introduction
1.1. Related Works
1.2. Contributions
2. Materials and Methods
2.1. Datasets
2.2. Two-Stage DNN Model Development
2.2.1. Image Pre-Processing
2.2.2. Data Augmentation
2.2.3. DNN Model
2.3. MRMC Observer Study Design
2.4. Statistical Analysis and Performance Evaluation
3. Results
3.1. Patient Characteristics
3.2. DNN Model Validation and Effect of Slice Thickness on Performance
3.3. MRMC Comparison Using an Independent Dataset
3.4. Cohen’s Kappa Statistic and Difference Significance Test
4. Discussion
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
Appendix A
Appendix B
Dataset | Manufacturer | Manufacturer Model Name | Convolutional Kernel | Number |
---|---|---|---|---|
Training Dataset (NCT = 1302) | Philips | Brilliance 64 | B | 59 |
L | 6 | |||
SIEMENS | SOMATOM Definition AS | B31f | 146 | |
B70f | 1 | |||
SOMATOM Definition AS | B31f | 135 | ||
B75f | 1 | |||
Sensation 40 | B31f | 1 | ||
Sensation 64 | B30f | 174 | ||
B31f | 718 | |||
B50f | 1 | |||
B70f | 59 | |||
TOSHIBA | Aquilion ONE | FC08 | 1 | |
Tuning Dataset (NCT = 365) | Philips | Brilliance 16 | B | 2 |
L | 293 | |||
TOSHIBA | Aquilion | FC51 | 1 | |
FC52 | 24 | |||
Aquilion ONE | FC51 | 34 | ||
FC52 | 8 | |||
FC86 | 2 | |||
United Imaging Healthcare | uCT 528 | B_SHARP_C | 1 | |
Validation Dataset 1 (NCT = 263) | GE MEDICAL SYSTEMS | LightSpeed VCT | BONEPLUS | 17 |
CHST | 88 | |||
STANDARD | 2 | |||
LightSpeed16 | BONEPLUS | 28 | ||
LUNG | 33 | |||
STANDARD | 44 | |||
Optima CT540 | BONEPLUS | 37 | ||
LUNG | 14 | |||
Validation Dataset2 (NCT = 175) | Philips | Brilliance 40 | C | 11 |
Ingenuity Flex | C | 1 | ||
YA | 1 | |||
iCT 256 | B | 38 | ||
L | 3 | |||
SIEMENS | SOMATOM Definition AS+ | B31f | 106 | |
United Imaging Healthcare | uCT 510 | B_SOFT_C | 11 | |
uCT 760 | B_SHARP_AB | 4 |
Model | Task1 | Task2 | ||||
---|---|---|---|---|---|---|
Ground Truth | Predicted Benign | Predicted Malignant | Ground Truth | Predicted Non-IA | Predicted IA | |
DNN | Benign | 49 | 30 | Non-IA | 83 | 3 |
Malignant | 20 | 103 | IA | 7 | 30 | |
Reader1 | Benign | 43 | 36 | Non-IA | 85 | 1 |
Malignant | 20 | 103 | IA | 19 | 18 | |
Reader2 | Benign | 30 | 49 | Non-IA | 62 | 24 |
Malignant | 9 | 114 | IA | 3 | 34 | |
Reader3 | Benign | 14 | 65 | Non-IA | 64 | 22 |
Malignant | 7 | 116 | IA | 2 | 35 | |
Reader4 | Benign | 59 | 20 | Non-IA | 85 | 1 |
Malignant | 59 | 64 | IA | 20 | 17 | |
Reader5 | Benign | 21 | 58 | Non-IA | 41 | 45 |
Malignant | 18 | 105 | IA | 0 | 37 | |
Reader6 | Benign | 30 | 49 | Non-IA | 61 | 25 |
Malignant | 29 | 94 | IA | 15 | 22 |
References
- Siegel, R.L.; Miller, K.D.; Jemal, A. Cancer Statistics, 2020. CA. Cancer J. Clin. 2020, 70, 7–30. [Google Scholar] [CrossRef]
- Aberle, D.R.; Adams, A.M.; Black, W.C.; Clapp, J.D.; Fagerstrom, R.M.; Gareen, I.F.; Gatsonis, C.; Marcus, P.M. Reduced Lung-Cancer Mortality with Low-Dose Computed Tomographic Screening—The National Lung Screening Trial Research Team. N. Engl. J. Med. 2011, 365, 395–409. [Google Scholar] [CrossRef] [PubMed]
- Travis, W.D.; Brambilla, E.; Noguchi, M.; Nicholson, A.G.; Geisinger, K.R.; Yatabe, Y.; Beer, D.G.; Powell, C.A.; Riely, G.J.; Van Schil, P.E.; et al. International Association for the Study of Lung Cancer/American Thoracic Society/European Respiratory Society International Multidisciplinary Classification of Lung Adenocarcinoma. J. Thorac. Oncol. 2011, 6, 244–285. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Ye, T.; Deng, L.; Wang, S.; Xiang, J.; Zhang, Y.; Hu, H.; Sun, Y.; Li, Y.; Shen, L.; Xie, L.; et al. Lung Adenocarcinomas Manifesting as Radiological Part-Solid Nodules Define a Special Clinical Subtype. J. Thorac. Oncol. 2019, 14, 617–627. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- MacMahon, H.; Naidich, D.P.; Goo, J.M.; Lee, K.S.; Leung, A.N.C.; Mayo, J.R.; Mehta, A.C.; Ohno, Y.; Powell, C.A.; Prokop, M.; et al. Guidelines for Management of Incidental Pulmonary Nodules Detected on CT Images: From the Fleischner Society 2017. Radiology 2017, 284, 228–243. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Hu, X.; Ye, W.; Li, Z.; Chen, C.; Cheng, S.; Lv, X.; Weng, W.; Li, J.; Weng, Q.; Pang, P.; et al. Non-Invasive Evaluation for Benign and Malignant Subcentimeter Pulmonary Ground-Glass Nodules (≤1 cm) Based on CT Texture Analysis. Br. J. Radiol. 2020, 93, 20190762. [Google Scholar] [CrossRef] [PubMed]
- Chae, H.-D.; Park, C.M.; Park, S.J.; Lee, S.M.; Kim, K.G.; Goo, J.M. Computerized Texture Analysis of Persistent Part-Solid Ground-Glass Nodules: Differentiation of Preinvasive Lesions from Invasive Pulmonary Adenocarcinomas. Radiology 2014, 273, 285–293. [Google Scholar] [CrossRef]
- Li, M.; Narayan, V.; Gill, R.R.; Jagannathan, J.P.; Barile, M.F.; Gao, F.; Bueno, R.; Jayender, J. Computer-Aided Diagnosis of Ground-Glass Opacity Nodules Using Open-Source Software for Quantifying Tumor Heterogeneity. Am. J. Roentgenol. 2017, 209, 1216–1227. [Google Scholar] [CrossRef]
- Mei, X.; Wang, R.; Yang, W.; Qian, F.; Ye, X.; Zhu, L.; Chen, Q.; Han, B.; Deyer, T.; Zeng, J.; et al. Predicting Malignancy of Pulmonary Ground-Glass Nodules and Their Invasiveness by Random Forest. J. Thorac. Dis. 2018, 10, 458–463. [Google Scholar] [CrossRef] [Green Version]
- Beig, N.; Khorrami, M.; Alilou, M.; Prasanna, P.; Braman, N.; Orooji, M.; Rakshit, S.; Bera, K.; Rajiah, P.; Ginsberg, J.; et al. Perinodular and Intranodular Radiomic Features on Lung CT Images Distinguish Adenocarcinomas from Granulomas. Radiology 2018, 180910. [Google Scholar] [CrossRef]
- Van Griethuysen, J.J.M.; Fedorov, A.; Parmar, C.; Hosny, A.; Aucoin, N.; Narayan, V.; Beets-Tan, R.G.H.; Fillion-Robin, J.C.; Pieper, S.; Aerts, H.J.W.L. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res. 2017, 77, e104–e107. [Google Scholar] [CrossRef] [Green Version]
- Fan, L.; Fang, M.J.; Li, Z.B.; Tu, W.T.; Wang, S.P.; Chen, W.F.; Tian, J.; Dong, D.; Liu, S.Y. Radiomics Signature: A Biomarker for the Preoperative Discrimination of Lung Invasive Adenocarcinoma Manifesting as a Ground-Glass Nodule. Eur. Radiol. 2018, 1–9. [Google Scholar] [CrossRef] [PubMed]
- Coudray, N.; Ocampo, P.S.; Sakellaropoulos, T.; Narula, N.; Snuderl, M.; Fenyö, D.; Moreira, A.L.; Razavian, N.; Tsirigos, A. Classification and Mutation Prediction from Non–Small Cell Lung Cancer Histopathology Images Using Deep Learning. Nat. Med. 2018, 24, 1559–1567. [Google Scholar] [CrossRef]
- Wang, S.; Zhou, M.; Liu, Z.; Liu, Z.; Gu, D.; Zang, Y.; Dong, D.; Gevaert, O.; Tian, J. Central Focused Convolutional Neural Networks: Developing a Data-Driven Model for Lung Nodule Segmentation. Med. Image Anal. 2017, 40, 172–183. [Google Scholar] [CrossRef] [PubMed]
- Ardila, D.; Kiraly, A.P.; Bharadwaj, S.; Choi, B.; Reicher, J.J.; Peng, L.; Tse, D.; Etemadi, M.; Ye, W.; Corrado, G.; et al. End-to-End Lung Cancer Screening with Three-Dimensional Deep Learning on Low-Dose Chest Computed Tomography. Nat. Med. 2019, 25, 954–961. [Google Scholar] [CrossRef] [PubMed]
- Zhao, W.; Yang, J.; Sun, Y.; Li, C.; Wu, W.; Jin, L.; Yang, Z.; Ni, B.; Gao, P.; Wang, P.; et al. 3D Deep Learning from CT Scans Predicts Tumor Invasiveness of Subcentimeter Pulmonary Adenocarcinomas. Cancer Res. 2018, 78, 6881–6889. [Google Scholar] [CrossRef] [Green Version]
- Wang, J.; Chen, X.; Lu, H.; Zhang, L.; Pan, J.; Bao, Y.; Su, J.; Qian, D. Feature-Shared Adaptive-Boost Deep Learning for Invasiveness Classification of Pulmonary Subsolid Nodules in CT Images. Med. Phys. 2020, 47, 1738–1749. [Google Scholar] [CrossRef]
- Gong, J.; Liu, J.; Hao, W.; Nie, S.; Zheng, B.; Wang, S.; Peng, W. A Deep Residual Learning Network for Predicting Lung Adenocarcinoma Manifesting as Ground-Glass Nodule on CT Images. Eur. Radiol. 2020, 30, 1847–1855. [Google Scholar] [CrossRef]
- Xia, X.; Gong, J.; Hao, W.; Yang, T.; Lin, Y.; Wang, S.; Peng, W. Comparison and Fusion of Deep Learning and Radiomics Features of Ground-Glass Nodules to Predict the Invasiveness Risk of Stage-I Lung Adenocarcinomas in CT Scan. Front. Oncol. 2020, 10, 418. [Google Scholar] [CrossRef]
- Gao, F.; Sun, Y.; Zhang, G.; Zheng, X.; Li, M.; Hua, Y. CT Characterization of Different Pathological Types of Subcentimeter Pulmonary Ground-Glass Nodular Lesions. Br. J. Radiol. 2019, 92, 20180204. [Google Scholar] [CrossRef]
- Son, J.Y.; Lee, H.Y.; Kim, J.-H.; Han, J.; Jeong, J.Y.; Lee, K.S.; Kwon, O.J.; Shim, Y.M. Quantitative CT Analysis of Pulmonary Ground-Glass Opacity Nodules for Distinguishing Invasive Adenocarcinoma from Non-Invasive or Minimally Invasive Adenocarcinoma: The Added Value of Using Iodine Mapping. Eur. Radiol. 2016, 26, 43–54. [Google Scholar] [CrossRef]
- Li, Q.; Fan, L.; Cao, E.T.; Li, Q.C.; Gu, Y.F.; Liu, S.Y. Quantitative CT Analysis of Pulmonary Pure Ground-Glass Nodule Predicts Histological Invasiveness. Eur. J. Radiol. 2017, 89, 67–71. [Google Scholar] [CrossRef]
- Gong, J.; Liu, J.; Hao, W.; Nie, S.; Wang, S.; Peng, W. Computer-Aided Diagnosis of Ground-Glass Opacity Pulmonary Nodules Using Radiomic Features Analysis. Phys. Med. Biol. 2019, 64, 135015. [Google Scholar] [CrossRef] [PubMed]
- Zhao, W.; Xu, Y.; Yang, Z.; Sun, Y.; Li, C.; Jin, L.; Gao, P.; He, W.; Wang, P.; Shi, H.; et al. Development and Validation of a Radiomics Nomogram for Identifying Invasiveness of Pulmonary Adenocarcinomas Appearing as Subcentimeter Ground-Glass Opacity Nodules. Eur. J. Radiol. 2019, 112, 161–168. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Wang, S.; Wang, R.; Zhang, S.; Li, R.; Fu, Y.; Sun, X.; Li, Y.; Jiang, X.; Guo, X.; Zhou, X.; et al. 3D Convolutional Neural Network for Differentiating Pre-Invasive Lesions from Invasive Adenocarcinomas Appearing as Ground- Glass Nodules with Diameters ≤3 cm Using HRCT. Quant. Imaging Med. Surg. 2018, 8, 491–499. [Google Scholar] [CrossRef] [PubMed]
- Wang, D.; Zhang, T.; Li, M.; Bueno, R.; Jayender, J. 3D Deep Learning Based Classification of Pulmonary Ground Glass Opacity Nodules with Automatic Segmentation. Comput. Med. Imaging Graph 2021, 88, 101814. [Google Scholar] [CrossRef]
- Wang, X.; Li, Q.; Cai, J.; Wang, W.; Xu, P.; Zhang, Y.; Fang, Q.; Fu, C.; Fan, L.; Xiao, Y.; et al. Predicting the Invasiveness of Lung Adenocarcinomas Appearing as Ground-Glass Nodule on CT Scan Using Multi-Task Learning and Deep Radiomics. Transl. Cancer Res. 2020, 9, 1397. [Google Scholar] [CrossRef] [PubMed]
- Hu, X.; Gong, J.; Zhou, W.; Li, H.; Wang, S.; Wei, M.; Peng, W.; Gu, Y. Computer-Aided Diagnosis of Ground Glass Pulmonary Nodule by Fusing Deep Learning and Radiomics Features. Phys. Med. Biol. 2021, 66, 065015. [Google Scholar] [CrossRef]
- Hu, J.; Chen, Y.; Yi, Z. Automated Segmentation of Macular Edema in OCT Using Deep Neural Networks. Med. Image Anal. 2019, 55, 216–227. [Google Scholar] [CrossRef]
- Chen, L.-C.; Papandreou, G.; Kokkinos, I.; Murphy, K.; Yuille, A.L. DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs. IEEE Trans. Pattern Anal. Mach. Intell. 2018, 40, 834–848. [Google Scholar] [CrossRef]
- Pedersen, J.H.; Saghir, Z.; Wille, M.M.W.; Thomsen, L.H.H.; Skov, B.G.; Ashraf, H. Ground-Glass Opacity Lung Nodules in the Era of Lung Cancer CT Screening: Radiology, Pathology, and Clinical Management. Oncology 2016, 30, 266–274. [Google Scholar] [CrossRef] [PubMed]
- Nemec, U.; Heidinger, B.H.; Anderson, K.R.; Westmore, M.S.; VanderLaan, P.A.; Bankier, A.A. Software-Based Risk Stratification of Pulmonary Adenocarcinomas Manifesting as Pure Ground Glass Nodules on Computed Tomography. Eur. Radiol. 2018, 28, 235–242. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Hao, P.; You, K.; Feng, H.; Xu, X.; Zhang, F.; Wu, F.; Zhang, P.; Chen, W. Lung Adenocarcinoma Diagnosis in One Stage. Neurocomputing 2019, 392, 245–252. [Google Scholar] [CrossRef]
- Hattori, A.; Hirayama, S.; Matsunaga, T.; Hayashi, T.; Takamochi, K.; Oh, S.; Suzuki, K. Distinct Clinicopathologic Characteristics and Prognosis Based on the Presence of Ground Glass Opacity Component in Clinical Stage IA Lung Adenocarcinoma. J. Thorac. Oncol. 2019, 14, 265–275. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Robbins, H.A.; Katki, H.A.; Cheung, L.C.; Landy, R.; Berg, C.D. Insights for Management of Ground-Glass Opacities from the National Lung Screening Trial. J. Thorac. Oncol. 2019, 14, 1662–1665. [Google Scholar] [CrossRef]
Characteristic | Training Dataset (n = 1476) | Tuning Dataset (n = 431) | Validation Dataset 1 (n = 284) | Validation Dataset 2 (n = 202) |
---|---|---|---|---|
Mean Age, y (SD) | 53.8 (±11.0) | 54.3 (±11.8) | 57.9 (±11.1) | 54.7 (±10.7) |
Sex, No. (%) | ||||
Male | 409 (31.4) | 129 (35.3) | 103 (39.2) | 57 (32.6) |
Female | 893 (68.6) | 236 (64.7) | 160 (60.8) | 118 (67.4) |
WHO pathological type, No. (%) | ||||
Benign/AAH | 206 (13.9) | 73 (16.9) | 38 (13.4) | 79 (39.1) |
AIS | 623 (42.2) | 77 (17.9) | 55 (19.4) | 53 (26.2) |
MIA | 261 (17.7) | 8 (1.9) | 64 (22.5) | 33 (16.3) |
IA | 386 (26.2) | 273 (63.3) | 127 (44.7) | 37 (18.3) |
Location, No. (%) | ||||
RUL | 543 (36.8) | 157 (36.4) | 118 (41.5) | 80 (39.6) |
RML | 110 (7.5) | 31 (7.2) | 17 (6.0) | 14 (6.9) |
RLL | 270 (18.3) | 76 (17.6) | 48 (16.9) | 27 (13.4) |
LUL | 384 (26.0) | 109 (25.3) | 71 (25.0) | 53 (26.2) |
LLL | 169 (11.4) | 58 (13.5) | 30 (10.6) | 28 (13.9) |
Nodule type on CT scan, No. (%) | ||||
pGGN | 1093 (74.1) | 308 (71.5) | 102 (35.9) | 175 (86.6) |
mGGN | 383 (25.9) | 123 (28.5) | 182 (64.1) | 27 (13.4) |
Diameter (mm), No. (%) | ||||
(3,10] | 888 (60.2) | 258 (59.9) | 135 (47.5) | 164 (81.2) |
(10,20] | 452 (30.6) | 143 (33.2) | 120 (42.3) | 36 (17.8) |
(20,30] | 136 (9.2) | 30 (6.9) | 29 (10.2) | 2 (1.0) |
Task1 | Task2 | ||
---|---|---|---|
Reference Standard of Diagnosis | Score | Reference Standard of Diagnosis | Score |
Highly suspicious normal/benign | 1 | Highly unlikely IA | 1 |
Moderately suspicious benign | 2 | Moderately unlikely IA | 2 |
Indeterminate/probably benign | 3 | Indeterminate | 3 |
Moderately suspicious malignant | 4 | Moderately suspicious IA | 4 |
Highly suspicious malignant | 5 | Highly suspicious IA | 5 |
Evaluation Metric | Task1 | Task2 | ||
---|---|---|---|---|
VD1 | VD2 | VD1 | VD2 | |
Accuracy (%) | 81.6 | 75.2 | 63.8 | 91.9 |
Sensitivity (%) | 90.7 | 83.7 | 86.6 | 81.1 |
Specificity (%) | 21.6 | 62 | 39.5 | 96.5 |
PPV (%) | 88.5 | 77.4 | 60.4 | 90.9 |
NPV (%) | 25.8 | 71 | 73.4 | 92.2 |
OR | 2.7 | 8.4 | 4.2 | 118.6 |
F1 (%) | 89.6 | 80.5 | 71.2 | 85.7 |
F1avg (%) | 80.9 | 74.9 | 61.6 | 91.7 |
MCC (%) | 13.2 | 47.1 | 29.7 | 80.3 |
Evaluation Index | Task1 | Task2 | ||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
DNN | R1 | R2 | R3 | R4 | R5 | R6 | DNN | R1 | R2 | R3 | R4 | R5 | R6 | |
Accuracy (%) | 75.2 | 69.8 | 71.3 | 64.4 | 60.9 | 62.4 | 61.4 | 91.9 | 83.7 | 78.0 | 80.5 | 82.9 | 63.4 | 67.5 |
Sensitivity (%) | 83.7 | 83.7 | 92.7 | 94.3 | 52.0 | 85.4 | 76.4 | 81.1 | 48.6 | 91.9 | 94.6 | 45.9 | 100.0 | 59.5 |
Specificity (%) | 62.0 | 54.4 | 38.0 | 17.7 | 74.7 | 26.6 | 38.0 | 96.5 | 98.8 | 72.1 | 74.4 | 98.8 | 47.7 | 70.9 |
PPV (%) | 77.4 | 74.1 | 69.9 | 64.1 | 76.2 | 64.4 | 65.7 | 90.9 | 94.7 | 58.6 | 61.4 | 94.4 | 45.1 | 46.8 |
NPV (%) | 71.0 | 68.3 | 76.9 | 66.7 | 50.0 | 53.8 | 50.8 | 92.2 | 81.7 | 95.4 | 97.0 | 81.0 | 100.0 | 80.3 |
F1 (%) | 80.5 | 78.6 | 79.7 | 76.3 | 61.8 | 73.4 | 70.7 | 85.7 | 64.3 | 71.6 | 74.5 | 61.8 | 62.2 | 52.4 |
F1avg (%) | 74.9 | 71.6 | 68.4 | 57.4 | 61.1 | 58.6 | 60.0 | 91.7 | 81.9 | 78.9 | 81.3 | 80.8 | 63.9 | 68.4 |
MCC (%) | 47.1 | 40.2 | 37.9 | 19.2 | 26.5 | 14.8 | 15.5 | 80.3 | 60.3 | 58.8 | 63.5 | 58.1 | 46.4 | 28.7 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Gong, J.; Liu, J.; Li, H.; Zhu, H.; Wang, T.; Hu, T.; Li, M.; Xia, X.; Hu, X.; Peng, W.; et al. Deep Learning-Based Stage-Wise Risk Stratification for Early Lung Adenocarcinoma in CT Images: A Multi-Center Study. Cancers 2021, 13, 3300. https://doi.org/10.3390/cancers13133300
Gong J, Liu J, Li H, Zhu H, Wang T, Hu T, Li M, Xia X, Hu X, Peng W, et al. Deep Learning-Based Stage-Wise Risk Stratification for Early Lung Adenocarcinoma in CT Images: A Multi-Center Study. Cancers. 2021; 13(13):3300. https://doi.org/10.3390/cancers13133300
Chicago/Turabian StyleGong, Jing, Jiyu Liu, Haiming Li, Hui Zhu, Tingting Wang, Tingdan Hu, Menglei Li, Xianwu Xia, Xianfang Hu, Weijun Peng, and et al. 2021. "Deep Learning-Based Stage-Wise Risk Stratification for Early Lung Adenocarcinoma in CT Images: A Multi-Center Study" Cancers 13, no. 13: 3300. https://doi.org/10.3390/cancers13133300