基于影像组学的多参数磁共振精准诊断前列腺癌病理Gleason分级的研究

Early diagnosis and treatment is the most effective way to improve the overall survival of patients with prostate cancer in China. The clinical decision of prostate cancer mainly depends on the pathological Gleason score at needle biopsy. However, about 1/3 of biopsies underestimate the aggressiveness of prostate cancer and these patients fail to get the optimal treatment at an early stage. Thus, it's urgent to develop a method to accurately diagnose the pathological Gleason score of prostate cancer. Multi-parameter magnetic resonance imaging is the best non-invasive method to diagnose prostate cancer at present, artificial intelligence has been proved to be valuable for predicting pathological grade of tumors. Our previous study has shown that radiomics could significantly improve the detection of prostate cancer in the gray zone of prostate specific antigen and also effectively predict the consistency of Gleason score between biopsy and radical prostatectomy. Based on our previous work, we plan to use the new technology of artificial intelligence to investigate the intrinsic relationship between magnetic resonance images and pathological Gleason score of prostate cancer, identify image signatures of high stability, high differentiation capability and high diagnostic accuracy, integrate them with clinical information to build predictive models, validate and optimize of models by using prospective data. We aim to realize noninvasive and precise diagnosis of pathological grade of prostate cancer, guide individualized treatment decision making of prostate cancer, and promote the clinical application of artificial intelligence technology.

早诊早治是提高我国前列腺癌患者总体生存率最有效的手段，现有前列腺癌的诊治决策主要取决于前列腺穿刺病理Gleason分级（GS），但约1/3的穿刺病理低估了前列腺癌的恶性程度，导致患者错失早期治疗的最佳时机，临床亟需研发早期精准诊断前列腺癌病理GS的方法。多参数磁共振是目前无创诊断前列腺癌的最佳方法，影像组学在预测肿瘤病理分级中的价值已得到初步证实。本项目组前期研究提示影像组学可提高对前列腺特异性抗原灰区内前列腺癌的检出，并可有效预测前列腺穿刺病理与根治切除术后病理GS的一致性。本项目拟在前期研究的基础上，借助人工智能新技术，进一步挖掘磁共振图像与前列腺癌病理GS之间的内在联系，筛选出高稳定性、高区分度、高诊断效能的影像标签，整合临床信息，构建预测模型，并在前瞻性数据上进行验证及优化，实现无创、精准诊断前列腺癌病理GS，辅助指导前列腺癌患者个体化的诊疗决策，促进人工智能技术的临床应用转化。

现有前列腺癌的诊治决策主要取决于前列腺穿刺病理Gleason分级（GS），但约1/3的穿刺病理低估了前列腺癌的恶性程度，临床亟需研发早期精准诊断前列腺癌病理GS的方法。多参数磁共振是目前无创诊断前列腺癌的最佳方法，影像组学在预测肿瘤病理分级中的价值已得到初步证实。针对目前GS早期精准诊断的临床难点，本项目提出了多参数MRI＋影像组学的解决方案，以期实现无创、精准前列腺癌病理Gleason分级预测，最终有助于指导前列腺癌的精准治疗。本项目最终构建并完善标准化前列腺癌多参数磁共振数据平台，基于数据平台研发前列腺癌Gleason分级预测模型、前列腺癌风险预测模型、前列腺癌包膜外浸润预测模型和前列腺自动分割模型各1个，并将该项目的影像组学和深度学习模型建立方法推广到其他泌尿系肿瘤的应用中，相关研究已发表SCI论著8篇，另有1篇SCI论著已接收，1篇在审稿中，1篇已投稿会议摘要，相关研究成果已提交专利申请2项。该项目培养青年科研骨干2人，新晋升高级职称1人，博士研究生2人，硕士研究生1人。该研究成果具有重要的临床转化意义，能为提高我国前列腺癌患者的生存率和降低死亡率提供科学依据，最终有助于促进重大研究计划总体目标的实现。该项目组多次派团队成员参与国际及国内知名学术会议，加深了团队与国际的学术交流，提高了我国学者在国际上的影响力。