The Application of Artificial Intelligence Technology in the Diagnosis of Acute Pancreatitis

Present acute pancreatitis surgical treatment often uses percutaneous catheter drainage minimally invasive surgery. Physicians generally based on the understanding of the patient's two-dimensional CT images to puncture and guide the catheter into position within the abdominal cavity to perform drainage of the free effusion to develop a surgical plan before surgery. The successful key to the CT-guided puncture and drainage is the accurate positioning of the inflammatory region. It is hard to divide the edges of each region accurately because there are many organs in abdomen CT image and some of the acute pancreatitis CT have free effusion regions. The localization and segmentation of each part in clinical treatment are influenced by the personal experience and ability of the physician. Computed Tomography Severity Index (CTSI) score is a result gotten from the analysis of the image to determine whether the patient has to get the operation or not. We collected 125 cases of CT images with the marked pancreas and free effusion area. Building models through Faster R-CNN object detection, the region growing and Gaussian filter to determine the survival pancreas, normal pancreas or free effusion area in CT images accurately. Reconstruct the three-dimensional structure image and related structure positioning, calculate the percentage of pancreatic necrosis, the number of necrotic areas and the size of the area, and then calculate the CTSI score. The experimental results were evaluated through three indicators of Dice Score, the volume overlap error (VOE) and the relative volume error (RVD). The accuracy rate of the automatic segmentation area compared to the standard area is 92 %, which can provide the physicians with an objective analysis of the surgical plan before surgery to improve the accuracy and safety.