Longitudinal Prostate MR Image Registration for Patients in Active Surveillance

1)简要介绍研究背景，包括潜在影响前列腺癌是西方世界许多地区男性诊断最多的癌症之一。作为最常见的观察策略之一，主动监测被用于帮助监测低风险前列腺癌患者的肿瘤进展，并决定接受治疗的时间点。挑战来自于接受标准的不确定性、决策的不确定性、随访时间间隔的确定以及患者的短期和长期预后。近年来，随着越来越多的患者加入到主动监测项目中，深度学习技术被认为可以帮助医生应对这些挑战。这既可以使医生从繁重的医学图像解读工作中解脱出来，又可以帮助医生做出更好的决策。2)目的和目标-具体目标是：对一个主动监测项目进行回顾性研究，建立基于深度学习的模型，帮助医生观察肿瘤的进展情况，制定治疗策略或决策。探索前列腺纵向图像配准方法，这是本课题的主要方法。项目推动的探索包括但不限于，例如研究患者间/患者内和单/多模态mri配准分布的影响。3)研究方法的新颖性：该方法专注于用弱监督的方式对纵向磁共振图像进行深度学习配准，可以在连续的随访中了解图像的形态变化。形态学变化可以被量化，并作为区分肿瘤进展状态的一个因素，这可以用于其他下游任务。4)与EPSRC的战略和研究领域保持一致：医疗技术：人工智能技术/医学成像5)涉及任何公司或合作者

1) Brief description of the context of the research including potential impactProstate cancer is one of the most diagnosed cancer in male in many parts of the western world. Asone of the most common observation strategies, active surveillance is used to help monitor the lowrisky prostate cancer patients on the progression of the tumor and to decide the time point to receivethe treatment. Challenges come from the uncertainties on accept criterion, decision making, determinethe time interval for follow-up visits and the short- and long-term outcomes of the patients. With theincreasing number of patients who were added into active surveillance project in recent years, deeplearning technologies are considered to use for helping the doctors face these challenges. Which mayboth relief the doctors form laborious medical image interpreting works and help make betterdecisions.2) Aims and Objectives-The specific objectives are to:1. Perform a retrospective study on an active surveillance project to build a deep learning basedmodel for helping the doctors observe the progression condition of the tumors and make treatmentstrategies or decisions.2. Exploring the method of longitudinal prostate image registration, which is a principle method inthis project. The project drives the explorations including but not limited to, for example, theresearch on the influence of distribution of the inter-/intra-patient and single/multi-modality MRimage registration.3) Novelty of Research MethodologyThe method is concentrating on the registration of the longitudinal magnetic resonance images withdeep learning in a weakly supervised manner, which could learn the morphological changes of theimages in a sequence of follow-up visits. The morphological changes could be quantified and be usedas a factor for discriminate the progression status of the tumor, which could be used in other downstreamingtasks.4) Alignment to EPSRC's strategies and research areasHealthcare Technologies: Artificial Intelligence Technologies / Medical Imaging5) Any companies or collaborators involvedNone.