HP Innovation Journal Issue 10: Fall 2018 - Page 71

we apply MVG-based methods directly, the reconstructed points no longer form a smooth surface, but rather represent scattered points spread out inside the cell. In other words, we need to reconstruct multiple structures inside the entire volume of the cell from the same number of images that surface reconstruction uses today. Our solution combined Structure from Motion (SfM) with predefined templates of cell structures. If we have prior knowledge about the cell, such as a few templates for the structures to be reconstructed, it is possible to reconstruct the entire cell volume from a finite number of key image features that defines each template. We then run the reconstruction algorithm on several video frames. Figure 3(a) shows a sample of the original images captured by the video camera. The reconstructed parametric digital model is shown in Figure 3(b). To visualize our model in the physical world, we designed a wire structure to see internal structures while preserving the shape of the cell membrane and nucleus. Figure 4 shows samples of 3D-printed cell models using the HP MultiJet Fusion 3D printing system. DISCUSSION Confocal microscopy has solved the problem of obtaining a 3D structure of a biological cell. In this solution, the sample is moved relative to imaging optics, while typically a pinhole in these optics ensures that the collected light originates from a single plane. Those planes are then combined to produce a 3D structure of the cell. This traditional solution has several complications. First, cost is typically high due to the need for precision stepping, confocal optics and high-sensitivity cameras. Second, there is the problem of data degradation. Light from distant parts of the sample must pass through sections of the sample proximal to the optics. And the proximal parts may act as a filter or add noise or interact with light, corrupting the imaging of the distant parts. Third, there is data uncertainty due to low photon flux, as data is collected from a thin plane. Our system costs significantly less for con- struction than traditional confocal systems because it does not require precision stepping, confocal optics and high sensitivity cameras. Additionally, the data we obtain is likely more accurate, as there is less transmission imag- ing—thus there is less of a filtering effect by parts of cells proximal to the imaging optics. To summarize, we built a noninvasive system that applies MVG to microscopic images of rotating cells, reconstructs 3D-cell models automatically and transforms them into our physical reality. In addition, as a system for cell visualization and modeling, our solution has broad applications, such as providing detailed information about unique cellular processes. We have applied our knowledge of microfluidics and computer vision to biology through live cell imaging. This innovation provides a quick and effective way to model cell structures. Also, by serving as a channel between the micro world that cells live in and our macro physical world, it has transformed cell geometry data to physical reality through 3D printing. Early Career Talents Innovation 71