The Construction Of AI Assisted Bronchoscopic Diagnosis Model Of Lung Cancer And Analysis Of Clinical Characteristics

Background and Purpose: Lung cancer is one of the malignant tumors with the highest morbidity and mortality in China and the world.Bronchoscopy is one of the most common diagnostic methods for lung cancer.In recent years,image recognition technology has been more and more widely studied and applied in the medical field,therefore,we constructed a bronchoscopic assisted diagnosis model of lung cancer based on deep learning method,we also analyzed the clinical characteristics of the666 cases of lung cancer.Method: A total of 666 patients with lung cancer and 152 patients with benign lesions were retrospectively obtained from the bronchoscopy center of the Third Xiangya Hospital of Central South University.The clinical data of patients were collected and statistically analyzed.The patients were divided into training set,verification set and test set according to 7:1:2,then we used the lesion images to construct a bronchoscopic assisted diagnosis model of lung cancer based on deep learning method.At the same time,9clinicians with different experience were invited to test and compare the diagnostic performance.Results: 1.In this study,male and smokers with lung cancer were significantly more than female and non-smokers.The highest incidence age group was 60-69 years old.Proliferative lesions were the most common type in squamous cell carcinoma,adenocarcinoma is dominated by invasive lesions,and there is no significant difference in the number of proliferative and invasive lesions in small cell lung cancer.The positive rate of SCC was the highest in squamous cell carcinoma,the CEA and CA125 in adenocarcinoma were the highest,and the positive rates of NSE and Pro GRP in small cell carcinoma were higher than other pathological types of lung cancer.Small cell lung cancer is easy to invade large blood vessels,and adenocarcinoma is most likely to have intrapulmonary metastasis;Small cell lung cancer and adenocarcinoma are more prone to lymphadenopathy and pericardial effusion than squamous cell carcinoma;squamous cell carcinoma and small cell carcinoma are more likely to develop obstructive pneumonia or atelectasis than adenocarcinoma.2.The overall accuracy,sensitivity,specificity and AUC of the benign and malignant differentiation model were 95.12%,97.76%,83.33% and 0.9396,which were equivalent to the judgment results of two doctors in the advanced group and higher than those of the other seven doctors.In the classification of squamous cell carcinoma and adenocarcinoma,the overall accuracy was 74.49%,including 79.03% for squamous cell carcinoma,66.67% for adenocarcinoma and the AUC value is 0.728.Conclusions: 1.In this study,the proportion of squamous cell carcinoma is the highest in patients with lung cancer,which is mainly composed of proliferative lesions.The changes of tumor markers were consistent with pathological changes.Small cell lung cancer and adenocarcinoma are most prone to multiple metastasis;2.We constructed the bronchoscopic assisted diagnosis model of lung cancer through image recognition technology;3 The diagnostic model can well distinguish benign and malignant lesions under bronchoscopy,it also can classify the pathological types of lung cancer to a certain extent.