Yagi, Y.; Groher, M.; Feuerstein, M.; Onozato, M.; Heibel, H.; Navab, N.
Overcoming Challenges in Histology 3D Imaging
Introduction: WSI technologies and rendering software have improved to the point that 3D reconstruction of large structures at microscopic scale from hundreds of serial sections became possible. 3D Imaging has the potential to bring about new discoveries in medicine. However, challenges remained such as section registration, quality of tissue and the effects of tissue processing and sectioning all must be optimized, and the huge amount of data that can be generated must be processed, stored and made available as quickly and efficiently as possible although we have been working on overcoming these issues for last several years, it was not easy and had limited usage of 3D imaging. Very recently, we improved the quality of consecutive image alignment technology and speed or reconstruction. It enhances the value of histology 3D imaging and opens up more possibilities. Methods: Specimens included mouse embryo, kidney, lung and stomach 50-250 serial sections were cut manually or by an automated sectioning machine (AS-200, KURABO INDUSTRIES LTD. Japan) from formalin-fixed paraffin-embedded blocks and stained with H&E. Serial sections were scanned at 0.33um/pixel using a Mirax Scan device (3DHISTECH Ltd, Hungary). 3D reconstruction was done using the algorithms developed by co-authors. To improve the quality of consecutive image alignment, new algorithms incorporated intensity values into the registration process underlying the image reconstruction. To this end, similarity measures working on pixel color values instead of extracted landmarks drive the iterative algorithm, which optimizes the relative geometric location between neighboring sections dramatically. To improve the reconstruction speed, new algorithm used a comhination of image pyramids and region processing: Image stacks were initially reeonstructed on a low magnification level, which did not cause too much computational workload. If a user selects a region of interest by zooming into the initially reconstructed volume, only the alignment of this particular region is updated discarding surrounding regions. Results: All 3D reconstruction results were improved from previous version. Previously when a case contains over 100 slides, we had to divide into multiple models due to the computer capacity. This time, 150-200 slides were reconstructed at once with the same computer. The improvement of the quality of consecutive image alignment gave us the opportunity to segment the region of interest and measure the volume and size of particular tissue components or organ and segmentation. Previously, we spent several hours to a couple of days to reconstruct one well aligned 3D model. However, with new technique it took 3-5 minutes for initial reconstruction and another 10 minutes for detailed reconstruction. Many usages of 3D imaging such as radiology view and blood vessel segmentation will be presented at the conference. Discussion: We believe that the technology for histology 3D imaging finally reached the level we expected many years ago. Next step is to improve the method to see the correlation between the 3D reconstructed images of histology slides with other 3D imaging modalities such as the microCT and optical frequency domain imaging (OFDI). The interface between different 3D imaging modalities has the potential to bring about new discoveries in medicine. Acknowledgement: Authors acknowledge to 3D Histech, Hungary and microDimensions GmbH