Purpose To develop an integrate model combining deep learning features from contrast-enhanced lung ultrasonography with clinical characteristics for predicting epidermal growth factor receptor mutation status in peripheral non-small cell lung cancer.Materials and Methods This retrospective study included 117 patients with pathologically confirmed non-small cell lung cancer from the First Affiliated Hospital of Guangxi Medical University(July 2021 to February 2024).Patients were randomly divided into training(n=93)and test(n=24)sets at an 8∶2 ratio.Regions of interest were delineated at the peak enhancement phase of contrast-enhanced lung ultrasonography.Various deep learning convolutional neural networks were pretrained,with ResNet18 selected as optimal for feature extraction.Deep learning,clinical,and integrated models were constructed using naive Bayesian algorithm.Performance was evaluated via receiver operating characteristic and calibration curves,while class activation mapping and Shapley additive explanation values provided model interpretability.Results In the training set,the deep learning,clinical and integrated models achieved area under the curve of 0.93(95%CI 0.88-0.98),0.86(95%CI 0.68-1.00),and 0.91(95%CI 0.85-0.97),respectively.Corresponding test set area under the curve were 0.81(95%CI 0.72-0.90),0.56(95%CI 0.33-0.80),and 0.87(95%CI 0.72-1.00).Both deep learning and integrated models significantly outperformed the clinical model in training(Z=2.380,P=0.017;Z=2.597,P=0.009)and test sets(Z=2.034,P=0.042;Z=2.577,P=0.010).The integrated model demonstrated excellent calibration and predictive performance.Conclusion The integrated model combining deep learning features from contrast-enhanced lung ultrasonography with clinical characteristics effectively predicts epidermal growth factor receptor mutation status in peripheral non-small cell lung cancer.

Purpose To develop an integrate model combining deep learning features from contrast-enhanced lung ultrasonography with clinical characteristics for predicting epidermal growth factor receptor mutation status in peripheral non-small cell lung cancer.Materials and Methods This retrospective study included 117 patients with pathologically confirmed non-small cell lung cancer from the First Affiliated Hospital of Guangxi Medical University(July 2021 to February 2024).Patients were randomly divided into training(n=93)and test(n=24)sets at an 8∶2 ratio.Regions of interest were delineated at the peak enhancement phase of contrast-enhanced lung ultrasonography.Various deep learning convolutional neural networks were pretrained,with ResNet18 selected as optimal for feature extraction.Deep learning,clinical,and integrated models were constructed using naive Bayesian algorithm.Performance was evaluated via receiver operating characteristic and calibration curves,while class activation mapping and Shapley additive explanation values provided model interpretability.Results In the training set,the deep learning,clinical and integrated models achieved area under the curve of 0.93(95%CI 0.88-0.98),0.86(95%CI 0.68-1.00),and 0.91(95%CI 0.85-0.97),respectively.Corresponding test set area under the curve were 0.81(95%CI 0.72-0.90),0.56(95%CI 0.33-0.80),and 0.87(95%CI 0.72-1.00).Both deep learning and integrated models significantly outperformed the clinical model in training(Z=2.380,P=0.017;Z=2.597,P=0.009)and test sets(Z=2.034,P=0.042;Z=2.577,P=0.010).The integrated model demonstrated excellent calibration and predictive performance.Conclusion The integrated model combining deep learning features from contrast-enhanced lung ultrasonography with clinical characteristics effectively predicts epidermal growth factor receptor mutation status in peripheral non-small cell lung cancer.