Prediction Model of Peripheral Pulmonary Lesions Based on R-EBUS Image

Diagnoses Disease

R-EBUS image, Deep learning, Peripheral pulmonary lesions, Prediction model, Diagnosis

Peripheral pulmonary lesions(PPLs) have a wide spectrum of diseases, and the diagnosis will affect the treatment strategy and prognosis. Radial endobronchial ultrasound (R-EBUS) can be used for non-invasive diagnosis of PPLs, and the supplement pathological diagnosis results of EBUS-TBLB, which has important clinical application value. This project intends to select representative images from R-EBUS dynamic videos for qualitative and quantitative analysis, to establish and verify the diagnostic evaluation system of R-EBUS forPPLs. Then build 1,000 R-EBUS image databases of PPLs, train deep learning networks for automatic extraction and diagnosis of target areas, and automatically extract representative images from videos to establish a benign and malignant prediction model of PPLs. We will provide reliable theoretical basis for the diagnosis of PPLs, and optimize the diagnosis and treatment method.The network would be prospectively verified through 300 R-EBUS images from multi centers.

PPLs are lesions at tertiary bronchus and above. The lesion cannot be seen by conventional bronchoscopy and the diagnosis will affect the treatment strategy and prognosis. R-EBUS can be used for non-invasive diagnosis of PPLs, and the supplement pathological diagnosis results of EBUS-TBLB. During the procedure, target PPLs are examined by ultrasound host (EU-ME2, Olympus, Tokyo, Japan) equipped with Doppler function and ultrasound probe . The bronchoscope reaches the distal as far as possible according to the predetermined position on chest CT or positron emission tomography-computed tomography (PET-CT) . The R-EBUS probe is inserted into the working channel of the bronchoscope, and gradually approaches the target PPL to obtain R-EBUS image. According to the characteristics such as within or adjacent to image, the probe scan the lesion from the near end to the far end and record the video. The recording time is required longer than 10 seconds. After selecting a typical R-EBUS image, freeze the image and take a screenshot. The long and short diameter of the lesion will be measured. This project includes three parts: preliminary construction and evaluation of R-EBUS image system for benign and malignant PPLs, construction of R-EBUS artificial intelligence prediction model and multi-center prospective validation of the prediction model. A total of 1000 patients will be enrolled to construct diagnostic model and 300 are enrolled to verify the diagnostic effiency.

No related papers found yet.

None listed

China