Region Detection Of Organs In Medical CT Images Based On Deep Learning

Medical Computed Tomography images with high resolution play a key role in the diagnosis of diseases.It is particularly becoming important to process and analysis of abdominal CT images because that Abdominal CT is an important means to check the health status of human organs.Abdominal CT images contain a large number of tissues and organs and the shape,size,and location of these organs are different in different abdominal CT images.In order to perform subsequent image processing and analysis such as edge extraction of the organ region,organ segmentation and organ three-dimensional reconstruction,it is necessary to perform the detection of the abdominal organ to determine the position of the organ in the CT image firstly.The method of manually marking the organ area by the medical staff is limited by the doctor's medical professional ability and subjective judgment.The existing region detection algorithms require to design the operators of feature extraction based on features of organs,which greatly limits the accuracy of region detection of multi-organs in medical CT images and the efficiency of region detection.In recent years,deep learning is becoming very popular and has achieved great success in the field of target detection because it relies on a deep learning model to learn features of images,so the main work of this article are as follows:Firstly,this paper proposes the use of existing target detection algorithm which is based on deep learning to determine the location of multiple organs in the abdominal CT image.And because of the scarcity of medical CT image sets,this paper uses data enhancement and transfer learning methods to train deep convolution neural networks under small samples.Secondly,this method is not accurate enough to locate the predicted region of the organ in the abdominal CT image,so combining the image connectivity to correct the predicted region.The effectiveness of the algorithm for region correction of organs is verified through experiments.Thirdly,because of the circular appearance of many organs in abdominal CT,an improved convolutional neural network target detection model was proposed based on previous work.The improvement is to change the rectangle bounding box generated in the algorithm to a circular bounding box which can reduce the redundant area in the detection of circular organ regions.In addition,the Inception structure is used to optimize the convolutional neural network structure of the target detection algorithm.The Inception structure can increase the width of the network structure to enhance the ability of network feature extraction and reduce the amount of computation.The experimental results also show that the improved convolutional neural network model can not only improve the recognition rate of circular-like organs and the overlap ratio between the prediction region and the real region,but also can improve the real-time detection of organ regions.