Selective plane illumination microscopy (SPIM) is becoming the method of choice for long-term 3D fluorescence imaging thanks to its low photo-toxicity and high imaging speed. However, SPIM is very data intensive: A single SPIM experiment can easily generate terabytes of image data, which is often overwhelming for biologists to handle. Moreover, large SPIM datasets often require additional computational power for
processing. There is a lack of optimized analysis software to visualize and quantify such large datasets. As a result, the data size burden is limiting the accessibility of this immensely powerful technology.
In this thesis, I investigated the root of the data burden in SPIM. I found that although the raw data volume generated by SPIM is large, the data product after processing is often very small in comparison. As a result, there are two ways of alleviating the data burden: the raw data to data product conversion ratio can be improved and the process of converting raw data to data product can be streamlined. In my Ph.D. project,
I demonstrated three different approaches to tackle the data burden in SPIM.
Firstly, I tested different lossless data compression methods on standard SPIM datasets that I collected during my thesis work. I found that integer compression algorithms are ideal for SPIM images. The data size can be reduced by more than half when compressed losslessly on-the-fly, reducing storage stress. Secondly, if the image quality could be improved, raw images would contain a higher amount of useful information and it would be less wasteful to store the large dataset. To illustrate this concept, I created a custom on-the-fly image analysis software that automatically selects the optimal imaging view in a multi-view SPIM experiment. By
applying the workflow to zebrafish embryo imaging, I showcase that each multi-view dataset contains more information than in the conventional case. Moreover, it became possible to reduce the number of imaging views without compromising data quality.
Lastly, raw data can also be converted into data product on-the-fly. The need to store raw images is often a result of the disconnect between imaging and data analysis. If the raw data can be analyzed in memory as soon as they are captured by the microscope, there is no need to keep the raw image data. I have built a custom image analysis pipeline to quantify zebrafish Rohon-Beard cells’ axon branching patterns. The image
analysis software semi-automatically performs sample surface extraction and image unwrapping. The resultant dataset is a flat lateral view of the embryo. The processed dataset is less than 10% the size of the original images. I also show that through directionality analysis, the processed data can be used to identify wild type embryos from drug-treated samples.
I also showcase a couple of other custom SPIM imaging workflows that I helped create. I have imaged patient-derived cancer spheroids and Xenopus oocytes in collaboration with other researchers. Here, the smart microscopy concept helped facilitate many data processing challenges involved. Overall, my thesis showcased that the data burden in SPIM can be addressed effectively by integrating image processing closely into the image capture process. I call this overall concept 'smart microscopy' and I believe it is the future of fluorescence microscopy.
| Identifer | oai:union.ndltd.org:DRESDEN/oai:qucosa:de:qucosa:37999 |
| Date | 24 January 2020 |
| Creators | He, Jiaye |
| Contributors | Grill, Stephan, Eliceiri, Kevin, Huisken, Jan, Technische Universität Dresden |
| Source Sets | Hochschulschriftenserver (HSSS) der SLUB Dresden |
| Language | English |
| Detected Language | English |
| Type | info:eu-repo/semantics/publishedVersion, doc-type:doctoralThesis, info:eu-repo/semantics/doctoralThesis, doc-type:Text |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0016 seconds