Research On A Non-Invasive Diagnostic Technique For Lung Cancer Combining Imaging And Exhaled Breath Markers

As one of the most common malignant tumors,lung cancer is a constant threat to human health.Early detection and treatment of lung cancer play an important role in improving the survival rate and quality of life of patients,but according to current research,it is still relatively difficult to confirm the diagnosis of lung cancer at an early stage,especially to achieve a more accurate judgment under the premise of non-invasive.More effective clinical methods are needed to achieve early diagnosis of lung cancer.PURPOSE:Computed tomography(CT)is a widely used method to diagnose benign lesions and lung cancer.Although analysis of CT scan results to aid in diagnosis has been clinically shown to be effective,it is limited by low sensitivity and reliance on subjective radiologist descriptions.Breath analysis(BA)is a promising non-invasive method for the diagnosis of benign lesions and lung cancer that utilizes volatile organic compounds(VOCs)found in human exhaled breath to diagnose disease and metabolic abnormalities.To address these limitations and improve the sensitivity and objectivity of diagnosis,this paper develops a model that combines Natural Language Processing(NLP)-based clinical text analysis(TA)of CT images with BA.METHODS:In this study,we collected breath samples from subjects by Proton Transfer Reaction Time Of Flight Mass Spectrometry(PTR-TOF-MS)and CT report text from subjects by CT,and finally collected 140 patients with malignancy and 91 lung patients with benign lesions for a total of 231 samples.Subjects were randomly assigned to the training and validation sets in a 4:1 ratio.the "what you see on examination" part of the CT report was used for training to develop the TA model.On the other hand,the BA model was developed using the Extreme Gradient Boosting(XGBoost)algorithm by cross-sectional comparison of three algorithms using VOCs in the subjects’ exhaled breath as predictors.The prediction results of the BA model and TA model were then used to develop a combined model based on XGBoost.RESULTS:Compared with the radiologist’s diagnosis(68.1%accuracy,74.3%sensitivity,59.1%specificity),the combined model showed better sensitivity and specificity,with 87.7%accuracy,94.3%sensitivity,and 77.3%specificity in the validation set,respectively.Meanwhile,the TA model had an accuracy of 75.4%,sensitivity of 74.3%,and specificity of 77.3%,and the BA model had an accuracy of 79.0%,sensitivity of 88.6%,and specificity of 63.6%.Comparison of the above evaluation indexes showed that the predictive ability of the combined model was better than that of the single BA model and the single TA model.Therefore,the combined model provides a more accurate and sensitive method for the diagnosis of benign and malignant lung lesions.