尿道高频超声因众多优点，有望满足尿道狭窄精准诊断与个体化治疗对影像技术的迫切需求。我们在前期研究中构建了节段性尿道超声三维影像，并基于此设计ULTRA评价体系，成功用于尿道狭窄术前精准诊断与术后疗效预测，得到国际同行认可。但受限于现有超声的影像重建依赖图像目标特征的人工提取，耗时长、效率低、精度差，且提取重建的三维影像为节段性，缺乏有效精准融合方法，严重限制该技术在临床推广应用。针对此问题，本项目拟采深度卷积神经网络算法进行尿道超声影像自动分析，并针对该影像自动分析的难点，聚焦于在前期研究中有重要突破的U-net模型的重建与优化，采用创新的密集连接残差U-net方法，通过对大样本图像学习，建立具有较强尿道超声图像特征学习和表达性能的模型，实现计算机辅助尿道超声图像自动分割和精准节段融合，解决快速化尿道三维影像重建和融合等关键科学问题，为尿道三维超声诊断技术应用临床化提供理论模型与关键技术。

Due to the numerous advantages, high-frequency ultrasound of the urethra is expected to meet the urgent need for imaging technology for accurate diagnosis and individualized treatment of urethral stricture. We constructed a segmental urethral ultrasound three-dimensional image in the previous study, and based on this design ULTRA evaluation system, it was successfully used for accurate diagnosis and postoperative efficacy prediction of urethral stricture, and was recognized by international peers. However, image reconstruction limited by existing ultrasound relies on manual extraction of image target features, which is time-consuming, inefficient, and inaccurate, and the extracted reconstructed 3D image is segmental, lacking effective and accurate fusion methods, severely limiting the technology in clinical practice. Promote the application. In response to this problem, this project plans to adopt the deep convolutional neural network algorithm for automatic analysis of urethral ultrasound images, and focuses on the reconstruction and optimization of U-net models with important breakthroughs in the previous research. Innovative densely connected residual U-net method, through the study of large sample images, establish a model with strong urethral ultrasound image feature learning and expression performance, to achieve computer-aided urethral ultrasound image automatic segmentation and precise segment fusion, to solve the rapidization Key scientific issues such as urethral three-dimensional image reconstruction and fusion provide theoretical models and key technologies for the clinical application of urethral three-dimensional ultrasound diagnosis technology.