| 研究実績の概要 |
1.While surgical videos are crucial for training young surgeons, traditional video capturing methods often suffer from significant occlusions caused by the movements of surgeons' heads and hands. Hence the surgical field, which is important for the transmission of procedural skills, is often invisible in the videos. To solve this problem, we proposes a multi-view capturing system formed by a ring-shape camera array. The occlusion can be avoided by shifting the viewpoints between cameras views. We also inducted bullet-time video technology to realize smooth and intuitive camera switching. Clinical experiments in the operating room are conducted to verify the effectiveness of the capturing method against the occlusion problem.
2.Intraoperative fluoroscopy is a frequently used modality in minimally invasive orthopedic surgeries. Aligning the intraoperatively acquired X-ray image with the preoperatively acquired 3D model of a computed tomography (CT) scan reduces the mental burden on surgeons induced by the overlapping anatomical structures in the acquired images. This paper proposes a fully automatic registration method that is robust to extreme viewpoints and does not require manual annotation of landmark points during training. It is based on a fully convolutional neural network (CNN) that regresses the scene coordinates for a given X-ray image. The scene coordinates are defined as the intersection of the back-projected rays from a pixel toward the 3D model.
|
| 今後の研究の推進方策 |
Radiography, a prevalent technique for visualizing internal human anatomy, employs high-energy X-rays to penetrate the body. The residual radiation energy is captured on a flat detector. Since various organs absorb X-rays differently, the measured energy is translated into a two-dimensional image, known as a radiograph or X-ray image, which discloses the body's internal configuration and offers crucial diagnostic data. X-ray images are both rapid and cost-effective; however, the use of high-energy radiation may have detrimental health implications.
Typically, in procedures like chest radiography, only a single frontal view is obtained per session. Although physicians can instinctively interpret the spatial arrangement of organs on a 2D radiograph in a three-dimensional context, this intuition is inherently subjective and can vary in precision.
The development of an X-ray image view synthesis algorithm could be beneficial, providing additional insight into a patient's internal anatomy. Furthermore, this could facilitate other applications, including sparse-view CT reconstruction from X-ray images, bridging the gap between CT and X-ray and enhancing the utility of radiographic imaging.
Diffusion models have shown extremely high performance in image generation tasks. It is also reported that with proper fine-tuning, a pre-trained model can generate realistic medical X-ray images from given text prompts. We hereby consider that with a carefully designed conditioning approach, it is possible to generate novel view X-ray images from a source image and a target viewport.
|
