Lung tumor detection based on improved Convolutional neural network
At present,Lung tumor is one of the diseases with the highest Case fatality rate rate.Its initial shape is generally very small and similar to normal tissue,and even experienced doctors cannot guarantee the precise location of the tumor.Therefore,using computer-aided detection is a good choice.On the basis of studying the deep learning object detection algorithm YOLOv5,aiming at the above difficulties,YOLOv5 algorithm is improved from the following three as-pects to detect Lung tumor.Firstly,the initial detection box size was redesigned based on the size of all tumors in the dataset;Then,the feature maps in the network were extracted using the backbone features,and a new detection layer was added.The feature maps of this detection layer were only downsampled twice,which can better preserve the detailed information of the tumor;Finally,the CBAM attention mechanism was selected to join the FPN structure to further im-prove the detection performance of the model.Through experiments on the LUNA16 dataset,it was found that the accuracy,recall,and mAP of the improved algorithm reached 96.81％,94.94％,and 96.6％,respectively,which were 15.18％,18.02％,and 13.46％higher than be-fore the improvement.Moreover,compared with similar algorithms in the past three years,it also has good detection performance.So this improved algorithm can effectively detect Lung tumor.
NETL
NSTL
万方数据
