Intraoperative integration of artificial intelligence during cystoscopic surgery

项目概要 膀胱癌是美国第六大常见癌症，是所有癌症中复发率最高的癌症之一 癌症，是从诊断到死亡治疗费用最高的癌症。膀胱现行标准 癌症诊断依靠临床白光膀胱镜检查进行初步筛查，然后进行经尿道检查 在手术室切除膀胱肿瘤进行病理诊断和局部分期。白光 膀胱镜检查有几个众所周知的缺点，特别是检测不完整，从而导致 次优切除并导致癌症复发和进展。我们的目标是改善结果 通过整合深度学习算法来改善膀胱镜检测，为膀胱癌患者提供治疗 并加强手术切除。 基于深度神经网络的人工智能（AI）已表现出卓越的学习能力 复杂的关系并将现有知识纳入推理模型。我们假设人工智能- 膀胱肿瘤的增强检测将改善临床环境中的诊断性膀胱镜检查，以识别 可疑病变，提高手术室经尿道电切术的质量，从而减少 总体癌症复发和结果。朝着建立基于人工智能的框架范式的目标 为了增强膀胱癌的检测，我们将利用我们强大的初步数据和杰出的 人工智能研究环境。我们提出三个具体目标：1）策划高质量的带注释膀胱镜检查 用于优化深度神经网络 CystoNet 的成像数据集； 2）实时设计和优化CystoNet 膀胱镜导航和癌症检测； 3) 进行前瞻性多中心验证 膀胱癌手术期间的 CystoNet。 成功完成此处提出的研究将有助于将深度学习算法转化为 膀胱镜手术的动态环境，无需专门的仪器。我们预见 我们的方法将改善主要癌症的治疗结果，并可推广到其他适合的器官系统 用于内窥镜干预。

PROJECT SUMMARY Bladder cancer is the sixth most common cancer in the U.S., has one of the highest recurrence rates of all cancers, and is the most expensive cancer to treat from diagnosis to death. Current standard for bladder cancer diagnosis relies on clinic-based white light cystoscopy for initial screening, followed by transurethral resection of bladder tumor in the operating room for pathologic diagnosis and local staging. White light cystoscopy has several well recognized shortcomings, particularly incomplete detection, thereby leading to suboptimal resection and contributing to cancer recurrence and progression. Our goal is to improve outcomes for bladder cancer patients through integration of a deep learning algorithm to improve cystoscopic detection and enhance surgical resection. Artificial intelligence (AI)-based on deep neural networks have demonstrated remarkable capacity to learn complex relationships and incorporate existing knowledge into the inference model. We hypothesize that AI- augmented detection of bladder tumor will improve diagnostic cystoscopy in the clinic setting to identify suspicious lesions and improve the quality of transurethral resection in the operating room, thereby reducing overall cancer recurrence and outcome. Towards the goal of establishing a paradigm of AI-based framework for augmented detection of bladder cancer, we will leverage our strong preliminary data and outstanding environment in AI research. We propose three specific aims: 1) To curate a high-quality annotated cystoscopy imaging dataset to optimize deep neural network CystoNet; 2) To design and optimize CystoNet for real-time cystoscopic navigation and cancer detection; and 3) To conduct a prospective multicenter validation of CystoNet during bladder cancer surgery. Successful completion of the studies proposed here will serve to translate deep learning algorithm to the dynamic environment of cystoscopic surgery without the need for specialized instrumentaitons. We foresee our approach will improve the outcome of a major cancer and genearlizable to other organ systems amenable for endsocopic interventions.