In recent years, we've focused on the OME analysis system and developing robust general image analysis methodology, culminating in our pattern recognition tool called WND-CHRM. We have validated this pattern-recognition approach to biological image analysis using diverse imaging modalities ranging from fluorescence microscopy to X-rays of human knees. We have also validated a range of applications from scoring image-based assays to diagnosis of disease to prediction of future disease risk. The specific applications of this approach are covered in reports AG000674-07 and AG000685-04. A major effort in the previous year has been to expand the functionality of WND-CHRM into real-time image-comparison assays as well as analysis of anatomically-defined patterns. A major effort is underway to rewrite the WND-CHRM code-base to make it more modular, better organized, easier to use, and accessible with the Python scripting language. WND-CHARM is a generalized pattern-recognition algorithm that can be used to analyze any type of image. Unlike most approaches to image processing in current use, this method relies on training a machine classifier to automatically recognize differences between training image classes (i.e. controls), rather than relying on an a-priori model of what is being imaged. This approach has been demonstrated to be effective at discerning differences even when they cannot be easily perceived manually. The output of a trained machine classifier is qualitative: for a given test image, it reports the most similar training class. In a scientific setting, it is often not sufficient to know what class an image belongs to, but how similar it is to the given training classes. An example is a quantitative imaging assay where the set of training image-classes comprise a standard curve, and the classifier's task is to arrive at a continuous score by interpolating between the defined classes. This type of classification can be called an """"""""ordered-class problem"""""""". There also exist a set of problems where the classes have no inherent order, and instead of an interpolated continuous score, the desired output is a measure of the similarity between classes. A familiar visualization of classes that have varying degrees of similarity to each other is a dendrogram for example, a phylogenetic tree representing evolutionary distance. This type of classification can be called a """"""""class-similarity problem"""""""". The current implementation of WND-CHARM addresses both of these quantitative imaging problems automatically. In addition to reporting the qualitative class assignment, it reports a continuous value if the class names can be interpreted numerically, and it computes pair-wise similarities between all of the classes. If a dendrogram visualization package is installed on the system (PHYLIP), it automatically generates a dendrogram based on the pair-wise class-distance matrix. This type of visualization has proven useful as an independent validation for ordered-class problems, since a well-ordered set of classes will produce a linear or elongated dendrogram without major branch-points. The program that implements the WND-CHRM algorithm (called wndchrm), has been made publicly available on Google Code (http://wnd-charm.googlecode.com/). A major release of the code (version 1.31) covering the areas discussed above has been made available on the project's site, as well as the Python code that is still under development. This release represents a first pass at reorganizing the code-base by making it more self-consistent and reliable without major architectural changes. It also represents a substantial effort in validation, testing and resolution of bugs. The site provides an interface for reporting bugs and requesting new features, and we have made extensive use of this facility within our own group. This continues to be visited multiple times per day, and the software source code has been downloaded several hundred times from multiple sites around the world. Whole-image analysis has proven very useful, but it is not always possible to compare whole images to each other. Examples of relatively homogenous images are those of cultured cells, or tissues like muscle, liver, and certain types of tumors. Our work on human knee X-Rays was the first application where a certain degree of pre-processing was necessary to make images of different subjects comparable to each other. In this case, we simply found the center of the knee joint in each image and extracted a fixed radius around this center for all patients. A much more complicated alignment problem exists in images with complex anatomy. Possibly the most extreme example of this are stained sections of brain tissue. A solution to the alignment problem would allow the use of generalized pattern recognition to address morphological differences in an anatomical context. For example, what areas of the brain correlate with cognitive decline or age? What is the degree of overlap between these areas? Spatially-resolved pattern analysis places an extreme burden on the performance of our software. Instead of an entire image being considered at once, or split into a small number of tiles on a grid, to achieve spatial resolution, each image must be sampled thousands or millions of times. In order to make this type of application practical, the computational strategy used in the software must be reconsidered. Previously, all 3,000 low-level image features were calculated for each image sample, even when most of them were later found to be irrelevant to the classification problem because they lacked discrimination power. The major change in strategy to enable spatially-resolved pattern recognition is to eliminate unnecessary calculations. This requires an on-demand computing strategy for image features, which is a major architectural goal for the wndchrm software. Recently, we have wrapped the C++ classes defined by wndchrm to make them accessible from Python. Python is a scripting language that is seeing increasing use in high-performance computing for manipulating large datasets. Currently, there is an operational prototype of this software that can compute image features on-demand. The software is publicly available under the """"""""pychrm"""""""" branch on our public code repository (http://code.google.com/p/wnd-charm/). Current efforts are underway to parallelize the execution of this code at a deeper level than is currently done in order to take better advantage of modern CPUs with multiple cores. The majority of our software-development efforts recently have been dedicated to the WND-CHRM analysis tool. With the addition of quantitative and spatially-resolved pattern analysis, it represents a substantial portion of what is possible with image analysis without a-priori models. Meanwhile, the OMERO project has matured under the guidance of Jason Swedlow, and is now a good, stable and usable implementation of the image and meta-data management concepts within OME. OMERO provides interface libraries for Python, which will facilitate its integration with pychrm. In the coming year, we plan a substantial effort of integration of WND-CHRM and OMERO.
| Du, Wei; Cheung, Huey; Goldberg, Ilya et al. (2015) A Longitudinal Support Vector Regression for Prediction of ALS Score. IEEE Int Conf Bioinform Biomed Workshops 2015:1586-1590 |
| Masuzzo, Paola; Martens, Lennart; 2014 Cell Migration Workshop Participants et al. (2015) An open data ecosystem for cell migration research. Trends Cell Biol 25:55-8 |
| Smith, Barry; Arabandi, Sivaram; Brochhausen, Mathias et al. (2015) Biomedical imaging ontologies: A survey and proposal for future work. J Pathol Inform 6:37 |
| Eliceiri, Kevin W; Berthold, Michael R; Goldberg, Ilya G et al. (2012) Biological imaging software tools. Nat Methods 9:697-710 |
| Orlov, Nikita V; Eckley, D Mark; Shamir, Lior et al. (2012) Improving class separability using extended pixel planes: a comparative study. Mach Vis Appl 23:1047-1058 |
