Introduction The new BTS Pulmonary Nodule Guidelines 2015 recommend the use of composite prediction models to assess the pre-test probability of malignancy in patients presenting with pulmonary nodules (PNs). These models were not developed for use in patients with a history of malignancy within five years of presentation with a PN.
In order to assist in the diagnosis of PNs, CT texture analysis has been proposed as a potential biomarker in tumour characterisation.1 Image texture refers to the statistical analysis of spatial intensity variations of the pixels within an image to produce a CT texture score.2
Aims and objectives
To evaluate four existing models for the probability of malignancy in the target population.
To create and validate prediction models for probability of malignancy for patients undergoing oncology follow-up for an indeterminate PN.
Methods Retrospective data on clinical and radiological characteristics were collected from the medical records of 61 patients with a PN (mean diameter 7 mm, SD 4 mm) that had an active or previous history (within 5 years) of primary lung or extra-thoracic malignancy. The gold standard diagnosis of the nodules was established by histology or 2-year stable follow-up.
Three multivariable logistic regression models were evaluated using a leave-one-out cross-validation strategy:
Model 1: Age, Sex, Smoking status, Emphysema, Nodule diameter.
Model 2: Age, Sex, Smoking status, Emphysema, CT Texture score.
Model 3: CT Texture score only.
The models’ performance, measured using the area under the ROC curve (AUC), were reported and further compared to existing clinical models.
Results The highest AUC, 0.86, was obtained from Model 3 (texture score only). Utilising clinical parameters (Model 2) did not improve performance.
In comparison, AUCs for previously published clinical models were 0.76(Mayo), 0.84(Herder), 0.66(VA) and 0.70(McWilliams) (Figure.1).
Conclusion This texture feature model is successful at discriminating benign from malignant nodules in a population of patients undergoing oncology follow-up.
While not significantly better than the Herder model (which incorporates PET avidity), this model offers improved risk stratification for PNs in the absence of PET in this patient group.
References 1 RSNA 2014, SSC03-05
2 IEEE International Conference doi: 10.1109/SMC.2013.663
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.