立体定向放射治疗(SBRT)是实现肿瘤精准治疗的重要手段,前列腺癌SBRT具有较好的临床应用前景。然而，由于盆腔器官的蠕动，以及膀胱直肠的充盈状态的变化，均可能导致在放疗分次内前列腺运动以及分次间前列腺位置的变化。对前列腺靶区内植入金属标记物进行实时肿瘤定位的方式具有侵入性创伤。目前仍没有文献报道精度较高的无创的前列腺癌运动位置变化的实时监测方法。本研究拟采用U-net深度神经网络方法，通过带有标记的前列腺癌患者的高清CBCT图像，构建直肠、膀胱及前列腺的自动勾画模型。在此基础上，通过膀胱体积变化速率、电磁引导定位器和体表轮廓信息，结合分次治疗前列腺、直肠和膀胱的空间位置信息，构建基于深度神经网络LSTM的前列腺运动预测模型，能在无创条件下，准确地实时监测前列腺位置变化，为前列腺癌患者开展立体定向放射治疗奠定了坚实的基础。

Stereotactic radiotherapy (SBRT) is an important method to achieve the precise treatment of tumors, and the prostate cancer SBRT has a better clinical application prospect. However, due to the peristalsis of the pelvic organs and the changes in the filling state of the bladder and rectum, it may result in the changes of the position of the prostate in the inter-fraction and intra-fraction radiotherapy. The method of real-time positioning of metal marker implanted in the prostate target area has invasive trauma. At present, there is no real-time monitoring method for the position change of the noninvasive prostate cancer with high accuracy. This study intends to use the method of U-net deep neural network to construct automated drawing model of rectum, bladder and prostate through high definition CBCT images of patients with marked prostate cancer. On this basis, based on the bladder volume change rate, electromagnetic guided locator and body surface contour information, combined with the spatial location information of prostate, rectum and bladder by each fraction, a prostate motion prediction model based on LSTM depth neural network can be constructed. It can accurately monitor the changes of prostate position in real time under non-invasive conditions, and lay a solid foundation for the stereotactic radiotherapy for prostate cancer patients.