Machine Learning Oriented Classification Algorithm For Lung Cancer Combined With Pulmonary Embolism

The probability of pulmonary embolism in lung cancer patients is more than 20 times that in normal people,and the mortality rate increases by 2 to 8 times.Once pulmonary embolism occurs in lung cancer patients will make diagnosis and treatment more difficult and shorten the survival period.Therefore,early diagnosis and prevention in patients with lung cancer complicated with pulmonary embolism are crucial.Based on clinical data and CT image data,this paper uses machine learning and deep learning methods to explore the application value of artificial intelligence technology for the classification of lung cancer and lung cancer complicated with pulmonary embolism.Firstly,a database of patients with lung cancer complicated with pulmonary embolism in Yunnan Province was established.A multicentre retrospective study was used to develop a predictive model for lung cancer complicated with pulmonary embolism.To explore the risk factors of pulmonary embolism in patients with lung cancer in Yunnan Province.The parameters was provided for an artificial intelligence predictive model for in in patients with lung cancer complicated with pulmonary embolism is of great value.The main research in this paper is divided into two main areas.（1）Based on clinical data for early diagnosis of lung cancer combined with pulmonary embolism,1NN-XGBoost hybrid network approach was proposed.Based on a dataset of 1572 patients,including 1372 cases of lung cancer and 200 cases of lung cancer complicated with pulmonary embolism.It was trained and tested to analyse the risk factors for lung cancer combined with embolism in Yunnan Province on average.1NN was used to address missing data and cost-sensitive function was used to address data imbalance.This data includes the features of 21 risk factors,and six machine learning algorithms and three deep learning algorithms were compared.The Area Under Curve（AUC）of 1NN-XGBoost hybrid network can reach 0.97.（2）X-ray based classification task for lung cancer combined with pulmonary embolism.Systems based on deep learning and image processing assist in diagnostic and preventive decision making for many diseases.This paper compares several approaches of CNN-1,CNN-2,Mobile Netv1,Effient Net B0 and migration learning in automatic multi-classification of lung cancer images.We used the available dataset from Yunnan First People’s Hospital and compared normal X-ray images（160）,lung cancer（1218）and lung cancer complicated by pulmonary embolism（88）,and the imbalance between the three categories was striking.Therefore,we tried migration learning using fine-tuning,resulting in an accuracy of over 95%,a 6% improvement over the results without imbalance treatment.Fine-tuned Efficient Net proved to be the most cost-effective performance.The method can assist doctors in decision making and improve diagnostic efficiency.