Lung Nodule Detection Algorithm Based On Improved Mask R-CNN

According to the data released by the World Health Organization,lung cancer ranks first in terms of incidence rate and mortality among all cancers,posing a serious threat to human life and health.When the lung nodule has malignant lesions,it may evolve into lung cancer.Therefore,if the lung nodule can be found at an early stage,it can help doctors diagnose and treat lung cancer patients early and improve the treatment effect of lung cancer.With the development of computer tomography technology,the number of medical images that radiologists need to observe is increasing,and the morphological characteristics of lung nodules are relatively complex with the surrounding information.Radiologists only rely on the naked eye and experience to evaluate and diagnose them,which not only consumes a lot of time and energy,but also tends to cause problems such as missed diagnosis or misdiagnosis.Therefore,the early detection of lung nodules has certain challenges.In this paper,the deep learning method is used to deeply study and analyze the lung nodule detection model and algorithm.Optimize and improve the shortcomings of Mask R-CNN algorithm in the detection of lung nodules,so as to improve the detection sensitivity of the algorithm to lung nodules and reduce the number of false positive lung nodules detected by the algorithm.First,the feature extraction network in the Mask R-CNN algorithm is improved to a recursive structure,so that the network can repeatedly learn and purify the features of lung nodules,and use the feature distribution matrix to enhance the information of its output feature map,thus improving the detection ability of the algorithm on lung nodules.Then the region proposal network in Mask R-CNN algorithm is improved to generate bounding boxes according to the probability of the growth position of lung nodules in the lung CT image,so as to improve the detection speed of the algorithm for lung nodules.Finally,a multi-scale 3D CNN network is constructed in the Mask R-CNN algorithm to enable the algorithm to obtain the information between axial slices of lung CT images,that is,to focus on the characteristic information of lung nodules in three-dimensional space,thereby reducing the number of false positive lung nodules generated when the algorithm detects lung nodules.The contrast experiment on LUNA16 dataset shows that the improved Mask R-CNN algorithm proposed in this paper is superior to other contrast algorithms in the detection task of lung nodules,and its detection sensitivity has reached a higher level in the overall detection performance.It can assist doctors in early detection of lung nodules,thus improving the survival rate of patients,and has certain clinical application value.