RESUMO
INTRODUCTION: Positron emission tomography/computed tomography (PET/CT) has an established role in evaluating patients with lung cancer. The aim of this work was to assess the predictive capability of [18F]Fluorodeoxyglucose ([18F]FDG) PET/CT parameters on overall survival (OS) in lung cancer patients using an artificial neural network (ANN) in parallel with conventional statistical analysis. METHODS: Retrospective analysis was performed on a group of 165 lung cancer patients (98M, 67F). PET features associated with the primary tumor: maximum and mean standardized uptake value (SUVmax, SUVmean), total lesion glycolysis (TLG) metabolic tumor volume (MTV) and area under the curve-cumulative SUV histogram (AUC-CSH) and metastatic lesions (SUVmaxtotal, SUVmeantotal, TLGtotal, and MTVtotal) were evaluated. In parallel with conventional statistical analysis (Chi-Square analysis for nominal data, Student's t test for continuous data), the data was evaluated using an ANN. There were 97 input variables in 165 patients using a binary classification of either below, or greater than/equal to median survival post primary diagnosis. Additionally, phantom study was performed to assess the most optimal contouring method. RESULTS: Males had statistically higher SUVmax (mean: 10.7 vs 8.9; p = 0.020), MTV (mean: 66.5 cm3 vs. 21.5 cm3; p = 0.001), TLG (mean 404.7 vs. 115.0; p = 0.003), TLGtotal (mean: 946.7 vs. 433.3; p = 0.014) and MTVtotal (mean: 242.0 cm3 vs. 103.7 cm3; p = 0.027) than females. The ANN after training and validation was optimised with a final architecture of 4 scaling layer inputs (TLGtotal, SUVmaxtotal, SUVmeantotal and disease stage) and receiving operator characteristic (ROC) analysis demonstrated an AUC of 0.764 (sensitivity of 92.3%, specificity of 57.1%). CONCLUSION: Conventional statistical analysis and the ANN provided concordant findings in relation to variables that predict decreased survival. The ANN provided a weighted algorithm of the 4 key features to predict decreased survival. IMPLICATION FOR PRACTICE: Identification of parameters which can predict survival in lung cancer patients might be helpful in choosing the group of patients who require closer look during the follow-up.
RESUMO
BACKGROUND: Strategies targeted at the five pillars of sustainability (social, human, economic, ecological and environmental) can be used to improve sustainability of clinical or research practices in nuclear medicine. KEY FINDINGS: While the core principle of sustainability is ensuring depletion does not exceed regeneration, this manuscript considers the balance of benefits and detriments of artificial intelligence (AI) technologies across the five pillars of sustainability. Specifically, innovations such as AI, generative AI and digital twins could enhance sustainability. While AI has the potential to address social asymmetry and inequity to drive the social and human pillars of sustainability, there is potential for widening the equity gap. AI augmentation and generative AI present economic and environmental sustainability opportunities. Deep digital twins offers clinical and research benefits in economic, ecological and environmental sustainability pillars. CONCLUSION: AI, digital twins and generative AI offer potential benefits to sustainability in nuclear medicine. Despite the benefits, caution is advised because these technologies confront a number of challenges that could potentially threaten sustainability. IMPLICATIONS FOR PRACTICE: AI presents opportunities for improving sustainability of nuclear medicine practice although caution is recommended to avoid unintentional undermining of sustainability across the five pillars.
RESUMO
INTRODUCTION: Extravasation or partial extravasation of the positron emission tomography (PET) tracer negatively effects image quality in PET and the accuracy of the standard uptake value (SUV). A commercially available topical sensor has been validated using a number of metrics to characterise injection quality. This evaluation explores contributing factors for extravasation and refines metrics to predict extravasation based on the time-activity-curves (TAC) of the topical sensor device. METHODS: A multi-site, multi-national pooling of 18F FDG PET/CT data was undertaken with 863 patients from 6 sites in the USA and 2 sites in Australia. A number of dose migration metrics determined with topical application of Lara sensors were retrospectively analysed using conventional statistical analysis. Deeper insights into the complex relationship between variables was further explored using an artificial neural network. RESULTS: Extravasation was independently predicted by the time taken for the injection sensor counts to reach double the counts of the reference sensor (tc50), the normalised difference between injection and reference sensors counts at 4 min post injection (ndAvgN), or the ratio of injection sensor counts to reference sensor counts at the end of data collection (CEnd ratio). The algorithm developed using the artificial neural network produced 100% sensitivity and 100% specificity against grounded truth for detecting extravasation by weighting and scaling these 3 key metrics; CEnd ratio, ndAvgN and tc50. CONCLUSION: Partial extravasation of a PET dose is readily detected and differentiated using TAC metrics and these metrics could provide deeper insight into the impact of partial extravasation on image quality or quantitation. Further validation of key metrics developed in this study are recommended in a larger and more diverse cohort. IMPLICATIONS FOR PRACTICE: Partial extravasation undermines image quality and accuracy of quantitation, impacting efficacy of outcomes for patients. Characterisation of extravasation informs decision making to optimise protocol and procedure, enhancing patient outcomes. Awareness provides the opportunity for education and training to minimise impact. The information can be used to drive policy and regulations to support improved techniques in practice.
