Research On The Segmentation Method Of Prostate TRUS Image Based On Mask R-CNN

With the improvement of people's living standards,prostate disease is currently one of the biggest threats to the health of male groups.Transrectal ultrasound images(TRUS)have the advantages of real-time,cheap and non-destructive,and are widely used in the prevention,diagnosis and treatment of prostate diseases.Especially in the treatment and diagnosis of prostate cancer,the demand for clinical applications such as biopsy needle placement is increasing,and fast and accurate TRUS image segmentation has become an indispensable technical requirement.Because the traditional manual segmentation TRUS image segmentation method not only relies on the experience of the diagnosing doctor,but also the segmentation efficiency is low.Therefore,efficient and accurate TRUS image segmentation has become one of the keys to real-time precise surgery for prostate cancer.This paper proposes a TRUS image segmentation method TSNet(TRUS Segmentation Network)based on Mask R-CNN.The MFPN network is used to enhance the features by controlling the channel scale on the multi-scale feature map.In order to effectively use features at different scales,the network uses top-down lateral connections and predicts features at different scales.This process is repeated repeatedly to generate features with rich semantic features and fuse the features.In the RPN stage,the E-OHEM optimization algorithm and the new Anchor mechanism are proposed,and the Anchor ratio is changed according to the size statistics of the prostate target area in the TRUS image.Random sampling according to a certain proportion of positive and negative samples in the small batch sampling process,and using non-maximum suppression to remove regions of interest with high overlap,not only improves the generalization ability of the model,but also can make the prostate edge more accurate Positioning.The experimental results show that the TSNet network proposed in this paper effectively solves the interference problem of the prostate segmentation caused by the high similarity between the artifact area and the target area and the problem of the weak prostate boundary contour.Compared with the standard value,the average absolute distance of the key indicators produced is 0.332 mm,the Hausdorf distance is 1.106 mm,the similarity coefficient is 0.9810,the specificity is 0.9983,and the sensitivity is 0.9764.A good segmentation result is obtained,which proves that The effectiveness of TSNet presented in this article.