Research On The Prediction Of Drug Resistance Of Non-small Cell Lung Cancer Based On Deep Learning

Lung cancer is the malignant tumor with the highest morbidity and mortality at present.About 80%-85% of lung cancers are non-small cell lung cancer(NSCLC),and molecular targeted drugs are often used clinically as its standard treatment.As time goes on,NSCLC patients are prone to develop drug resistance,and the occurrence of drug resistance varies from person to person,which increases the difficulty of clinical treatment.Tumor is dynamic biological system that changes over time,which means that some information can be extracted from its change process to help analyze the occurrence of drug resistance in patients.In recent years,deep learning has been rapidly developed,and it has also been successfully applied in the filed of medical image processing.Based on the theory of Convolutional Neural Networks(CNN)and Recurrent Neural Networks(RNN),this paper conducts research on the prediction of NSCLC resistance.The main work are as follows:1.Collecting CT images data of NSCLC patients at different time points after drug treatment.These data include images of tumor parts,thus they can form a set of time series images which reflecting tumor changes.2.Construct a deep learning model based on CNN and RNN to predict the type of drug resistance in patients by analyzing time series images of tumors.Using multiple CNNs to extract tumor features at different time points,and input these features into RNN for further longitudinal analysis,so as to output a prediction result which reflecting the type of tumor resistance.Divide the data set into a training group and a test group at a ratio of 3:1,and use transfer learning to complete the construction of the model3.Evaluate the classification performance of the model on the test set,and compare the impact of different numbers of CNNs on the model performance when processing time series images.Experimental results show that as the number of CNN models that process time series data increases,the classification results of the models can be significantly improved.When using three CNNs,the model can achieve an accuracy of 78.95% on the test set and an AUC value of 0.81,which indicate that the method is feasible to predict the type of drug resistance in NSCLC patients.