2005 Scientific Session Abstracts
Virtual Placenta: Computational Phenotyping Through Image Analysis
Tony Pan, MS 1 (tpan@bmi.osu.edu); Kishore Mosaliganti, MS 2; Richard Sharp, BS 2; Randall Ridgway, BS 2; Kun Huang, PhD; Raghu Machiraju, PhD 2; Joel Saltz, MD, PhD 1; 1Biomedical Informatics, Ohio State University, Columbus, OH; 2Department of Computer Science and Engineering, Ohio State University, Columbus, OH
Context : Light microscopic imaging is an important tool for characterizing tissue morphology and pathology. Structural and functional changes due to disease, genetic modification, and pharmaceutical manipulation can be digitally assessed via microscopy. Here we present our effort to computationally phenotype fetal tissue morphology changes in mouse placentas with retinoblastoma (Rb) gene knocked out.
Technology : The mouse placenta is sectioned at 3 micron thickness and stained with hematoxylin and eosin. The serial sections are scanned using Aperio ScanScope slide digitizer and uncompressed images are stored on a Linux cluster. Two mouse placentas have been scanned with approximately 2000 images and 1.7 terabytes of data. Algorithms were developed using the Visualization Toolkit (VTK), Insight Segmentation and Registration Toolkit (ITK), MatLab, and C++. Images are processed on Linux clusters.
Design : Reconstruction of the serial sections into a 3D virtual placenta is accomplished by registering successive images. Mutual information based registration process was developed using VTK and ITK. Different fetal tissue types are segmented using a Bayesian supervised segmentation method developed in MatLab. 3D surfaces are then constructed from the placenta and the tissue layers, using VTK and an in-house developed implicit surface reconstruction algorithm. The 3D placenta and tissue layers are visualized using VTK, Ohio Supercomputing Center’s VolSuite, and a custom volume rendering tool which emphasizes tissue layer boundaries. Volume and surface area are computed for the different tissue layers.
Results : We constructed a computational phenotyping pipeline for registering, segmenting, surface reconstructing, and visualizing genetically modified mouse placentas. A mutant placenta and a wild type placenta were processed using this pipeline. Intermediate results are presented for each of the steps.
Conclusions : We have developed capabilities to quantitatively assess phenotypic changes in 3D microscopy datasets. We have applied the pipeline to analyze genetically modified mouse placentas.