Research And Implementation Of Pneumonia Detection Method Based On Deep Learning

About 4 million people die each year from pneumonia globally,and is still increasing year by year.Early diagnosis of pneumonia can effectively reduce mortality.Medical resources are extremely scarce in many areas in China,and the advent of a pneumonia-assisted diagnosis system can help alleviate this phenomenon.Pneumonia detection is one of the most crucial steps in the pneumonia diagnosing system,clinical information of patients plays an important role in the detection of pneumonia.The main contributions of this paper are as follows:First,a lung parenchyma segmentation algorithm based on improved region growing algorithm is proposed.This method can automatically obtain seed points,and then growing based on the seed points to achieve lung parenchyma segmentation.the multi-plane hole filling algorithm is used to repair lung parenchyma.Second,A model(PDNet)using CNN and LSTM is proposed.The method uses three-channel CT images,patient age and gender to simulate the clinical pneumonia detection process.In order to reduce the need for computing resources,the CT sequence is regarded as a short video frame.First,a parameter-sharing CNN is used to extract features in all CT images,and then use LSTM to extract the contact information in these features.Finally,PDNet uses the patient’s age and gender information as a priori to improve the model’s performance.Third,In order to prove the effectiveness of the model,using 1002 clinical cases provided by the Southwest Hospital,comparative experiments and ablation experiments were given for experimental analysis.The accuracy of the proposed model is 0.940,and it has a very balanced performance in terms of sensitivity and specificity.The method proposed in this paper proves that multi-modal data provides more sufficient information than image data alone,and obtains more convincing results.