1. Thoracic imaging radiomics for staging lung cancer: a systematic review and radiomic quality assessment
- Author
-
Alexander C. Simone, Isabella F. Churchill, Waël C. Hanna, Forough Farrokhyar, Anthony A. Gatti, Kerrie A. Sullivan, Grigorios I. Leontiadis, and Yogita Patel
- Subjects
medicine.medical_specialty ,business.industry ,medicine.medical_treatment ,MEDLINE ,Evidence-based medicine ,medicine.disease ,Malignancy ,Radiation therapy ,Medical imaging ,medicine ,Radiology, Nuclear Medicine and imaging ,Observational study ,Radiology ,Lung cancer staging ,Lung cancer ,business - Abstract
Radiomics, a method used to extract large amount of data, may be useful for discriminating malignant characteristics for lung cancer staging. We aimed to critically appraise the current use of radiomics in medical imaging for determining nodal involvement in lung cancer. The literature was searched using Cochrane Central Register of Controlled Trials, MEDLINE, EMBASE and Web of Science for observational studies between the databases’ inception and January 2020. All observational studies compared diagnostic tests for lung cancer in comparison to diagnostic tests with a radiomic adjunct. For each study, data were extracted for study characteristics, methodological quality, and accuracy. Discrepancies in screening and extraction were resolved by a third reviewer. Data were synthesized narratively. 15 studies were included, 13 of which were full reports, while two were conference abstracts. Overall, 3164 patients were enrolled with a total of 3379 LNs recorded (69% malignancy). Point estimate sensitivities for algorithms ranged from 52 to 99%, while specificity for the algorithms ranged from 60 to 92%. Area under the curve c-statistics were only reported for algorithms and not the clinician. Point estimate c-statistics ranged from 0.64 to 0.94, suggesting that the algorithms possessed good discrimination potentials. The current level of evidence of radiomic analysis methods for staging lung cancer is inconclusive and possesses heterogeneity in study design. Prospective external validation of these algorithms and direct comparisons using cut-off thresholds is required to determine their true predictive capability prior to implementation in clinical practice. PROSPERO ID: CRD42020162952.
- Published
- 2021
- Full Text
- View/download PDF