Lung Cancer Classification System Beased On PET/CT Fusion

The screening of lung cancer often needs PET-CT and other medical imaging auxiliary means.Different types of lung cancer should be treated by corresponding medical means.It is not only inefficient but also unsatisfactory to rely on the naked eye judgment of doctors.Lung cancer detection and discrimination is a very important task in the field of medical image processing.Many machine learning related theories and methods will be used.However,it is not easy to extract image features accurately and apply them to machine learning to solve complex problems.Facing the growing demand of cancer diagnosis and the task of lung cancer classification which is time-consuming,labor-consuming and low accuracy,this paper studies the possibility of lung cancer classification by neural network and designs different network structures for comparative experiments.In previous studies,the two modalities of PET＿CT were often used separately.In this paper,we make full use of multi-modalities,neural network to investigate the effect of CT and PET data separately as input,as well as the effect of neural network which uses different network structures to fuse CT and PET data.Three Fusion methods which contain fusion of pixel maps(FPM),fusion of feature maps(FFM),and fusion of feature fectors(FV)were compared in the system implementation process.In addition,SUV was introduced into the treatment of pet to further improve the classification effect.The experimental results show that the network that using fusion of CT and PET as input is better than any of single one;and different fusion methods have different degrees of accuracy improvement;in processing PET data,the introduction of SUV is better than process directly.Finally,the accuracy of the lung cancer classification task was 69.4%.