Research And Implementation Of Lung Image Segmentation Data Enhancement Algorithm Based On Model Learning Space And Density Conversion

With the rapid development of machine learning and deep learning technology,intelligent image processing technology has been widely used in the medical field.In many modern studies on lung diseases,CT(computed tomography)lung substantive segmentation is the core step of in-depth research.At present,among various methods of lung segmentation,supervised learning methods have achieved the best accuracy.However,the training of these supervised learning methods relies heavily on large data sets with labels.On the other hand,labeling CT scans is a very time-consuming task,and different data sets obtained from different types of devices may have various data characteristics,resulting in very different performance on different models.How to construct reliable and accurate labeled data from the existing data information as much as possible is a very challenging topic.The task of this subject is to design and implement a lung image segmentation data enhancement framework to facilitate the research work of relevant researchers.This paper first studies and analyzes the development of related fields,and proposes a single lung segmentation framework based on learning space and density conversion.The framework requires only a segmented CT scan(source scan)and uses spatial and density conversion to convert this scan according to the target scan.These target scans can be scans from another data set or scans from other types of devices.The convolutional neural network U-net is used to learn the spatial deformation field and structural density of the lung to achieve the conversion of space and density.Then,a supervised R-FCN will be trained to segment the lungs from the CT scan using the generated scan.In this paper,the one-shot method is used to segment the lung,and the method is experimentally demonstrated.Experiments show that even if the labeled source scan comes from another data set,the experimental results of lung image segmentation data enhancement have achieved a convincing effect.This paper introduces the design and implementation of this lung image segmentation data enhancement framework in detail.The implementation of the framework provides an effective and reliable thinking direction for relevant scientific researchers,and its powerful functionality is also easy to use,effectively promoting the development of related fields.