科研之友
微信
新浪微博
Facebook
分享链接
摘要
Three-dimensional representations in light microscopy are important for accurate shape assessment of model systems in biosciences. The computational multiview 3-D reconstruction seems feasible in obtaining the 3-D representations in particular for high-throughput. The specimen for imaging can have properties, i.e., transparency and translucency, that impede the detection of well-defined boundaries. Consequently, 3-D reconstruction and measurements, i.e., volume and surface area will be inaccurate. The motivation in this paper is therefore to develop a two-phase 3-D reconstruction approach for light microscopy axis-view imaging that can deal with these properties. In phase I of this approach, we develop an improved 3-D volumetric representation defined as the confidence map. This is derived from texture-augmented axial-view images of the specimen. In phase II, the 3-D reconstruction is accomplished by searching the optimal surface for the specimen over the confidence map. Subsequently, from the obtained 3-D reconstruction, 3-D measurements can be extracted. We present a high-throughput axial-view imaging architecture in light microscopy based on the vertebrate automated screening technology device. Using this imaging architecture, we present three typical datasets using different imaging modalities, which includes zebrafish larvae in bright-field and zebrafish liver in fluorescence. In the experiments, we have applied our approach on these datasets. We find that our approach yields precise 3-D shape representation and natural visualization. In comparison with a groundtruth setup, we have obtained accurate 3-D measurements both for the organism and the organ, which holds a promising shape assessment for model systems in biosciences.
- 出版日期2017-10
