Visentini-Scarzanella Marco, Sugiura Takamasa, Kaneko Toshimitsu, Koto Shinichiro
Multimedia Laboratory, Toshiba Corporate Research and Development Center, 1, Komukai-Toshiba-cho, Kawasaki, 212-8582, Japan.
Int J Comput Assist Radiol Surg. 2017 Jul;12(7):1089-1099. doi: 10.1007/s11548-017-1609-2. Epub 2017 May 15.
In bronchoschopy, computer vision systems for navigation assistance are an attractive low-cost solution to guide the endoscopist to target peripheral lesions for biopsy and histological analysis. We propose a decoupled deep learning architecture that projects input frames onto the domain of CT renderings, thus allowing offline training from patient-specific CT data.
A fully convolutional network architecture is implemented on GPU and tested on a phantom dataset involving 32 video sequences and [Formula: see text]60k frames with aligned ground truth and renderings, which is made available as the first public dataset for bronchoscopy navigation.
An average estimated depth accuracy of 1.5 mm was obtained, outperforming conventional direct depth estimation from input frames by 60%, and with a computational time of [Formula: see text]30 ms on modern GPUs. Qualitatively, the estimated depth and renderings closely resemble the ground truth.
The proposed method shows a novel architecture to perform real-time monocular depth estimation without losing patient specificity in bronchoscopy. Future work will include integration within SLAM systems and collection of in vivo datasets.
在支气管镜检查中，用于导航辅助的计算机视觉系统是一种有吸引力的低成本解决方案，可引导内镜医师针对周围病变进行活检和组织学分析。我们提出了一种解耦的深度学习架构，该架构将输入帧投影到CT渲染域上，从而允许根据患者特定的CT数据进行离线训练。
在GPU上实现了一个全卷积网络架构，并在一个包含32个视频序列和60k帧的模型数据集上进行测试，该数据集具有对齐的地面真值和渲染结果，这是作为支气管镜导航的第一个公共数据集提供的。
平均估计深度精度为1.5毫米，比从输入帧进行的传统直接深度估计高出60%，在现代GPU上的计算时间为30毫秒。定性地说，估计深度和渲染结果与地面真值非常相似。
所提出的方法展示了一种新颖的架构，可在支气管镜检查中进行实时单目深度估计，同时不丧失患者特异性。未来的工作将包括集成到SLAM系统中以及收集体内数据集。
