Research On Intelligent Screening And Diagnosis Algorithm Of Pulmonary Nodules Based On CT Images

Among the ten most common malignant tumors in the world,lung cancer ranks first with 18%mortality rate and second with 11.4%incidence rate.Lung cancer poses a huge threat to human life and health.Early screening,early diagnosis and early treatment are the most effective means to improve patient survival and reduce mortality.CT imaging is the most common imaging method used by clinicians to screen for early lung cancer,and pulmonary nodules are the most significant imaging manifestations of lung cancer.The method of clinical manual screening will increase the burden of reading images and work pressure for doctors,resulting in missed and misdiagnosed lung cancer screening results.The research on intelligent screening and diagnosis algorithm of pulmonary nodules based on CT images is of great significance for assisting doctors to save the lives of lung cancer patients.In view of this,taking lung CT images as the research object,combined with imaging features of pulmonary nodules,this paper carried out research on the intelligent detection algorithm of pulmonary nodules and the intelligent diagnosis algorithm of benign and malignant.The main research work is as follows:(1)Aiming at the weak automation level,low detection accuracy and slow detection efficiency of the intelligent lung cancer screening system,a pulmonary nodule detection algorithm that balances accuracy and efficiency is designed.Based on the YOLOv4 algorithm,the detection accuracy of lung nodules is improved by designing depthwise hyperparameterized convolutional layers,convolutional block attention module,and focal loss function.The redundant convolution kernel of the model is trimmed by the Network Sliming strategy to improve the detection efficiency.The experimental results on the LIDC-IDRI dataset show that the algorithm proposed in this paper has high detection accuracy and detection efficiency.(2)Aiming at the uneven distribution of lung CT image data and the high complexity of 3D convolutional neural network design,a 3D pulmonary nodule benign and malignant classification algorithm based on neural architecture search is designed.Based on the partial order pruning search strategy and relying on data-driven,the computer can adaptively build a network structure that best fits the distribution of CT image data;This paper also designs a cross-dimensional interactive Quadruple Attention module to increase the feature extraction and feature representation capabilities of the 3D classification network for pulmonary nodules.The experimental results on the LUNA 16 dataset show that the algorithm proposed in this paper has better classification accuracy and automation.(3)To simulate the ’expert consultation’ scheme of clinicians in lung cancer diagnosis,a multi-model classification result fusion decision-making method based on ensemble learning is designed to improve the reliability of diagnosis.In this paper,multiple models of neural architecture search are used as base classifiers,relying on the difference and complementarity of the features learned by different base classifiers,and the ensemble learning strategy is used to fuse the benign and malignant diagnosis results of multiple models.The experimental results on the LUNA 16 dataset show that the algorithm proposed in this paper has good classification accuracy and reliability.