Research On Prostate MR Image Segmentation Algorithm Based On Deep Learning

Prostate cancer is one of the most common male cancers worldwide,and Magnetic Resonance Imaging(MRI)plays a crucial role in the diagnosis and treatment of prostate diseases.However,the diverse morphology and fuzzy boundaries of prostate tissue pose many limitations and defects for traditional algorithms in prostate segmentation.In addition,the traditional manual segmentation method requires a lot of time and labor,and is prone to problems such as inconsistency and subjectivity.In recent years,deep learning-based medical image automatic segmentation technology has made significant progress.In this paper,we conducted research on prostate MR image segmentation based on deep learning,and the main contents and innovative work of the paper are as follows:1)In view of the problems of gradient vanishing and over-segmentation in traditional deep segmentation neural networks,a CC-Res-UNet network is proposed.Firstly,the Contrast-Limited Adaptive Histogram Equalization(CLAHE)algorithm is applied to process prostate images,enhancing the detectability of information.Then,by introducing the residual mechanism into the classical UNet network,the problem of gradient vanishing caused by deepening the neural network is alleviated.In addition,a correction module with classification guidance function is introduced between the encoder and decoder of the traditional segmentation network,in order to reduce false positive predictions and alleviate the over-segmentation problem of the segmentation network for images without prostate tissue in prostate MR images.2)To enhance the feature expression ability of the network and achieve the fusion of multi-scale semantic information,an Inception A-Dense-UNet network is proposed.First,the dense connection idea is introduced into the UNet model to improve the connection mode of the original encoder and decoder,and to achieve the fusion and propagation of multi-scale semantic information.In addition,the cascade convolution operation is replaced by the Inception module driven by the dilated convolution to increase the width of the network,enhance the feature extraction and expression ability for different size targets,and improve the segmentation performance of the network.3)To address the problem of data dependence in supervised learning in deep learning,a semi-supervised SCC-U2-Net network is proposed.Traditional deep learning methods require a large amount of annotated data for training,but the cost of acquiring annotated data is often high.This paper combines self-training methods with the idea of regular consistency to improve the segmentation performance and generalization ability of the model with only a small amount of labeled data,greatly reducing the dependence on labeled data.In the experimental section,the NCI-ISBI 2013 Challenge public dataset was used to verify the performance and to conduct comparative experiments with traditional segmentation networks.The results show that the data processing method used can reduce network training costs without compromising segmentation performance.Additionally,the data augmentation method used can generate more diverse synthetic images and improve the generalization performance of the segmentation model.Furthermore,the improved segmentation model outperforms traditional segmentation models in multiple indicators,demonstrating better accuracy and segmentation effects.Finally,the image segmentation method based on semi-supervised learning designed in this paper can use less labeled data for training and reduce dependence on labeled data.In summary,the prostate MR image segmentation method proposed in this paper based on deep learning has good application prospects and can provide effective assistance for the diagnosis and treatment of prostate diseases.