The emergence of deep learning-based convolutional neural networks has made breakthrough progress in target image recognition and detection,natural image semantic segmentation.This thesis discusses the detection and segmentation of the target area of medical images in detail,and deeply explores the correlations within the pathological images.The detection of the trigeminal nerve in the target cranial MR images region is one of the important basis for the diagnosis of trigeminal nerve disease in the brain.However,the common medical image detection algorithm is to detect or enhance the edge and contour of a tissue in the medical image.Few studies on the automatic detection of trigeminal nerve regions in cranial MR images.The traditional method of identifying trigeminal nerves is to manually find the craniocerebral level with trigeminal nerves in a set of cranial MR and calibrate its position by a doctor.This method is time-consuming and labor-intensive.Aiming at the above problems,this paper has made related research in real-time detection and extraction of trigeminal nerves in the craniocerebral.The segmentation of the rectal cancer tumor in the CT image of the abdominal transverse position has a certain reference value for the diagnosis and treatment of early rectal cancer,judging the invasion and metastasis of the lesion,and the clinical treatment of prognosis.Since the image data of rectal cancer tumors have complex lesion characteristics and relatively regular expression,high-quality annotation data is relatively small.How to segment the CT image of the abdominal transverse position in view of the above problems.And judging the relationship between tumor and lymph node metastasis of rectal cancer is another main research content of this article about the intelligent judgment of rectal cancer lymph nodeThe main content of this thesis is divided into two parts:1)The YOLO network was used to automatically detect the trigeminal nerve region of the cranial MRI image,and then the C-V model was used to realize the rapid segmentation of the trigeminal nerve.In order to improve the inference speed,and then,systematically evaluated the inference performance of the NVIDIA TensorRT framework under different computing platforms.2)A multi-scale rectal cancer tumor segmentation network model based on 3D CNN CT rectal biphasic tumor segmentation was designed for rectal cancer tumor segmentation with CT images of abdominal transverse position.This model used three-dimensional image input and multi-scale input of the two-phase small-sample data set of the arterial phase and portal phase of rectal CT to achieve rectal cancer tumor segmentation through three-dimensional convolution.and the segmented rectal cancer tumor image was used to analyze the correlation between tumor image and tumor lymph node metastasis.Simultaneously,the relationship between tumor histology features,overall image features and lymph node metastasis was analyzed,a classifier was established,the classification results were compared and analyzed,and the effectiveness of different classification models in predicting lymph node prediction of rectal cancer was evaluated.The experimental results of the above two studies showed that:1)YOLO target detection network provides a well initial contour basis for the automatic segmentation of the trigeminal neural region of the brain,and provides a reliable input image for the accurate segmentation and analysis of the trigeminal nerve based on the C-V model.simultaneously,under the framework of NVIDIA TensorRT,when the input brain MRI image resolution was(204x204),the optimized trigeminal neural target detection YOLOv2 network was available on the CPU platform,embedded GPU platform and desktop GPU platform and professional GPU computing card platform.the frame rates per second were 0.1FPS,23.4FPS,and 793.7FPS.This provides important reference for the subsequent development of portable trigeminal neural segmentation equipment.2)Multi-scale rectal cancer tumor segmentation under CT dual rectal stage based on 3D CNN has achieved good segmentation results.The final segmentation accuracy rate was more than 90%,which provided a strong basis for the judgment of lymph node metastasis.There is a strong advantage in pre-diagnosis.By comparing the prediction results of rectal cancer lymph nodes based on machine learning SVM,Logistics,decision tree,naive Bayes classifier and based on convolutional neural network,the classification model based on convolutional neural network performs best,which achieved on the test data set The recognition accuracy rate of 74.23%indicated that the features extracted by the convolutional neural network were rich.It can provide intelligent judgment for lymph node metastasis of rectal cancer. |