In the computer field, image is a common data type. Image preprocessing is the operation of image preprocessing before image processing and analysis. The purpose of this chapter is to take image preprocessing as an example to explore the influence of abdominal CT images before and after preprocessing on pancreatic segmentation. 3D-UNet network of convolutional neural network is used for image segmentation, and NIH public pancreas data set is used for data set. In the experiment, the abdominal CT images were preprocessed by affine transformation, elastic deformation, and random flip enhancement. The evaluation result uses the Dice similarity coefficient (DSC) to assess the similarity between the correct pancreas and the pancreas organ segmented through the experiment. Through the experiment, the DSC values of CT images without pretreatment and those pretreated by the above three methods are 73.63%, 76.01%, 74.28%, and 72.17%, respectively. It can be seen from the experimental data that, compared with the pancreas CT image without image preprocessing, the pancreas CT image with affine transformation and elastic deformation preprocessing has improved the segmentation accuracy by 2.38% and 0.65%, respectively, while the pancreas CT image with random flip enhancement preprocessing has reduced the segmentation accuracy by 1.46%. It can be concluded that whether the image is preprocessed or not does affect the pancreatic segmentation results of 3D-UNet, and different preprocessing methods have different effects on the segmentation results. These findings have certain guiding significance for improving the accuracy and reliability of medical image segmentation.