Comparison of 18F-FDG avidity at PET of benign and malignant pure ground-glass opacities: a paradox? Part II: artificial neural network integration of the PET/CT characteristics of ground-glass opacities to predict their likelihood of malignancy

Abstract

Aim: To assess the ability of artificial neural networks (ANNs) to predict the likelihood of malignancy of pure ground-glass opacities (GGOs), using observations from computed tomography (CT) and 2-[¹⁸F]-fluoro-2-deoxy-d-glucose (FDG) positron-emission tomography (PET) images and relevant clinical information.

Materials and methods: One hundred and twenty-five cases of pure GGOs described in a previous article were used to train and evaluate the performance of an ANN to predict the likelihood of malignancy in each of the GGOs. Eighty-five cases selected randomly were used for training the network and the remaining 40 cases for testing. The ANN was constructed from the image data and basic clinical information. The predictions of the ANN were compared with blinded expert estimates of the likelihood of malignancy.

Results: The ANN showed excellent predictive value in estimating the likelihood of malignancy (AUC = 0.98±0.02). Employing the optimal cut-off point from the receiver operating characteristic (ROC) curve, the ANN correctly identified 11/11 malignant lesions (sensitivity 100%) and 27/29 benign lesions (specificity 93.1%). The expert readers found 23 lesions indeterminate and correctly identified 17 lesions as benign.

Conclusion: ANNs have potential to improve diagnostic certainty in the classification of pure GGOs, based upon their CT appearance, intensity of FDG uptake, and relevant clinical information, and may therefore, be useful to help direct clinical and imaging follow-up.