Fei Kang1, Wei Mu2,3, Jie Gong4, Shengjun Wang1, Guoquan Li1, Guiyu Li1, Wei Qin5, Jie Tian6,7, Jing Wang8. 1. Department of Nuclear Medicine, Xijing Hospital, Fourth Military Medical University, No.127 West Changle Road, Xi'an, China. 2. Key Laboratory of Molecular Imaging of Chinese Academy of Sciences, Institute of Automation, Chinese Academy of Sciences, No. 95 Zhongguancun East Road, Beijing, China. 3. Department of Cancer Physiology, H. Lee Moffitt Cancer Center, Tampa, FL, USA. 4. Life Sciences Research Center, School of Life Sciences and Technology, Xidian University, No. 266 Xifeng Road, Xi'an, China. 5. Life Sciences Research Center, School of Life Sciences and Technology, Xidian University, No. 266 Xifeng Road, Xi'an, China. wqin@xidian.edu.cn. 6. Key Laboratory of Molecular Imaging of Chinese Academy of Sciences, Institute of Automation, Chinese Academy of Sciences, No. 95 Zhongguancun East Road, Beijing, China. jie.tian@ia.ac.cn. 7. Life Sciences Research Center, School of Life Sciences and Technology, Xidian University, No. 266 Xifeng Road, Xi'an, China. jie.tian@ia.ac.cn. 8. Department of Nuclear Medicine, Xijing Hospital, Fourth Military Medical University, No.127 West Changle Road, Xi'an, China. wangjing@fmmu.edu.cn.
Abstract
PURPOSE: The high false positive rate (FPR) of 18F-FDG PET/CT in lung cancer screening represents a severe challenge for clinical decision-making. This study aimed to develop a clinical-translatable radiomics nomogram for reducing the FPR of PET/CT in lung cancer diagnosis, and to determine the impact of integrating manual diagnosis to the performance of the radiomics nomogram. METHODS: Among 3,947 18F-FDG PET/CT-screened patients with lung lesion, 157 malignant and 111 benign patients were retrospectively enrolled and divided into training and test cohorts. The data of manual diagnosis were recorded. A total of 4,338 features were extracted from CT, thin-section CT, PET and PET/CT, and the four radiomics signatures (RS) were then generated by LASSO method. Radiomics prediction nomogram integrating imaging-based RS and manual diagnosis was developed using multivariable logistic regression. The performances of RS and prediction nomograms were independently validated through key discrimination index and clinical benefit. RESULTS: The FPR of manual diagnosis was found to be 30.6%. Among the four RS, PET/CT RS exhibited the best performance. By integrating manual diagnosis, the hybrid nomogram integrating PET/CT RS and manual diagnosis demonstrated lowest FPR and highest area under curve (AUC) and Youden index (YI) in both training and test cohorts (FPR: 5.4% and 9.1%, AUC: 0.98 and 0.92, YI: 85.8% and 75.5%, respectively). This hybrid nomogram respectively corrected 78.6% and 37.5% among FPR cases produced by PET/CT RS, without significantly sacrificing its sensitivity. The net benefit of hybrid nomogram appeared highest at <85% threshold probability. CONCLUSION: The established hybrid nomogram integrating PET/CT RS and manual diagnosis can significantly reduce FPR, improve diagnostic accuracy and enhance clinical benefit compared to manual diagnosis. By integrating manual diagnosis, the performance of this hybrid nomogram is superior to PET/CT RS, indicating the importance of clinicians' judgement as an essential information source for improving radiomics diagnostic approaches.
PURPOSE: The high false positive rate (FPR) of 18F-FDG PET/CT in lung cancer screening represents a severe challenge for clinical decision-making. This study aimed to develop a clinical-translatable radiomics nomogram for reducing the FPR of PET/CT in lung cancer diagnosis, and to determine the impact of integrating manual diagnosis to the performance of the radiomics nomogram. METHODS: Among 3,947 18F-FDG PET/CT-screened patients with lung lesion, 157 malignant and 111 benign patients were retrospectively enrolled and divided into training and test cohorts. The data of manual diagnosis were recorded. A total of 4,338 features were extracted from CT, thin-section CT, PET and PET/CT, and the four radiomics signatures (RS) were then generated by LASSO method. Radiomics prediction nomogram integrating imaging-based RS and manual diagnosis was developed using multivariable logistic regression. The performances of RS and prediction nomograms were independently validated through key discrimination index and clinical benefit. RESULTS: The FPR of manual diagnosis was found to be 30.6%. Among the four RS, PET/CT RS exhibited the best performance. By integrating manual diagnosis, the hybrid nomogram integrating PET/CT RS and manual diagnosis demonstrated lowest FPR and highest area under curve (AUC) and Youden index (YI) in both training and test cohorts (FPR: 5.4% and 9.1%, AUC: 0.98 and 0.92, YI: 85.8% and 75.5%, respectively). This hybrid nomogram respectively corrected 78.6% and 37.5% among FPR cases produced by PET/CT RS, without significantly sacrificing its sensitivity. The net benefit of hybrid nomogram appeared highest at <85% threshold probability. CONCLUSION: The established hybrid nomogram integrating PET/CT RS and manual diagnosis can significantly reduce FPR, improve diagnostic accuracy and enhance clinical benefit compared to manual diagnosis. By integrating manual diagnosis, the performance of this hybrid nomogram is superior to PET/CT RS, indicating the importance of clinicians' judgement as an essential information source for improving radiomics diagnostic approaches.
Authors: Geoffrey D Rubin; John K Lyo; David S Paik; Anthony J Sherbondy; Lawrence C Chow; Ann N Leung; Robert Mindelzun; Pamela K Schraedley-Desmond; Steven E Zinck; David P Naidich; Sandy Napel Journal: Radiology Date: 2004-11-10 Impact factor: 11.105
Authors: Bruno Hochhegger; Matheus Zanon; Stephan Altmayer; Gabriel S Pacini; Fernanda Balbinot; Martina Z Francisco; Ruhana Dalla Costa; Guilherme Watte; Marcel Koenigkam Santos; Marcelo C Barros; Diana Penha; Klaus Irion; Edson Marchiori Journal: Lung Date: 2018-10-09 Impact factor: 2.584
Authors: Hyungjin Kim; Chang Min Park; Jeonghwan Gwak; Eui Jin Hwang; Seon Young Lee; Julip Jung; Helen Hong; Jin Mo Goo Journal: AJR Am J Roentgenol Date: 2018-11-26 Impact factor: 3.959
Authors: Philippe Lambin; Emmanuel Rios-Velazquez; Ralph Leijenaar; Sara Carvalho; Ruud G P M van Stiphout; Patrick Granton; Catharina M L Zegers; Robert Gillies; Ronald Boellard; André Dekker; Hugo J W L Aerts Journal: Eur J Cancer Date: 2012-01-16 Impact factor: 9.162
Authors: Mubarik A Arshad; Andrew Thornton; Haonan Lu; Henry Tam; Kathryn Wallitt; Nicola Rodgers; Andrew Scarsbrook; Garry McDermott; Gary J Cook; David Landau; Sue Chua; Richard O'Connor; Jeanette Dickson; Danielle A Power; Tara D Barwick; Andrea Rockall; Eric O Aboagye Journal: Eur J Nucl Med Mol Imaging Date: 2018-09-01 Impact factor: 9.236
Authors: Arthur Jochems; Turkey Refaee; Henry C Woodruff; Philippe Lambin; Guangyao Wu; Abdalla Ibrahim; Chenggong Yan; Sebastian Sanduleanu Journal: Eur J Nucl Med Mol Imaging Date: 2021-03-11 Impact factor: 10.057