Ultrasound-based Diagnostic Algorithm Design For Pneumothorax

Pneumothorax is the most common injury in war trauma environment and disaster emergency rescue scene.Pneumothorax can cause difficulty in breathing and even threaten the life of the wounded in severe cases.Early diagnosis and treatment are urgently needed.The application of ultrasound in the diagnosis of pneumothorax has gradually matured.Due to the characteristics of light weight and small size of ultrasound equipment,it has been widely used in war trauma environments and disaster emergency rescue sites.However,physicians using ultrasound require long-term training,and the results of diagnosis are often affected by the user’s skill level.A large number of studies have shown that pattern recognition methods such as machine learning are gradually playing a key role in the fields of medical imaging and computeraided diagnosis.While improving the recognition accuracy,the diagnosis efficiency has been greatly improved.The application in the field of war trauma environment and disaster emergency rescue The advantages are obvious.In this paper,an ultrasound-based pneumothorax diagnosis algorithm is proposed.Taking the portable and miniaturized embedded ultrasound equipment as the research background,in order to make the algorithm suitable for embedded chips,based on five types of lung ultrasound images under M-mode ultrasound,compressed sensing technology is used for M-mode ultrasound images.Data compression is performed on M-mode ultrasound images,and a single-hidden layer feedforward neural network with a simple structure is used to identify and classify the compressed signals.(1)This paper selects 5 types of lung ultrasound images for classification,of which2 types of images represent the signs that appear in normal lungs,2 types of images represent the signs that appear when the lungs produce pneumothorax,and type 1represents the signs that are occluded by ribs.Due to the lack of ultrasound pneumothorax data,in order to facilitate the acquisition of a large number of ultrasound pneumothorax images,a lung simulation model made of water-absorbent resin and water bags is designed in this paper to simulate the structure of the lung and to simulate the two types of abnormalities when pneumothorax occurs sign.(2)Explain the principle and application steps of the compressed sensing algorithm in detail,and use the compressed sensing technology to compress the data of five types of ultrasound images.At the same time,the compressed ultrasound image is reconstructed with the reconstruction algorithm of compressed sensing to verify the availability of the compressed data.(3)Use a single hidden layer feedforward neural network to train the compressed data,and use another data set to identify and classify the trained single hidden layer network.At the same time,two kinds of commonly used deep learning network models are set to use the same training set and test set for training,recognition and classification.The classification and recognition effects of deep learning network model and singlehidden layer feedforward network model on ultrasound images are discussed from the aspects of accuracy,algorithm complexity,and floating-point computation.