Long-term goals
The goal of this project is to advance the state of the art in automated cell and tissue image analysis by developing an associative image analysis system named FARSIGHT (Fluorescence Association Rules for Multi-dimensional Insight) that is specifically designed for image-based measurement needs of complex tissue-level units. It will be disseminated freely to researchers via the Internet.
Our expectations from FARSIGHT are the following:
• Ability to handle complex problems: It will be possible to quantify objects and relationships in complex microenvironments involving multiple cell types and vasculature using this toolkit.
• Programmability: The tools will be adaptable to a new application with the minimum of training. They will be free from inherent limitations that may prevent broader application in the future.
• High level of automation, exceeding 95%, and designed to minimize the manual burden associated with initial setup, supervision, validation, performance assessment, oversight, and corrective editing.
• Simplicity & Clarity: The system will be based on a conceptually clean and simple framework that guides software developers and biologists alike. This will enable biologists to plan experiments knowing what to expect on the image analysis side. At the same time, software developers will have a clear sense of what problems to work on.
• Modularity: The framework will be modular and allow extensions to be incorporated in the future. There will be no inherent structural limitations to prevent such expansion.
• Robustness: The software algorithms must be robust to common image degradations such as noise, blur, imaging artifacts, and biological variability across specimens within a study. At the same time, users will have a sense of just how robust the algorithms are, so they can plan their experiments accordingly.
• Practical Parallel Processing: The tools will be written from the outset to be free from inherent limitations that prevent parallel processing. Specifically, they should be capable of exploiting low-cost parallel processing capabilities already available in the current generation of multi-core desktop computers, laboratory servers, and supercomputers when available. When possible, we will use graphics processing units (GPUs).
• Scalability: It should be possible to scale up the automated image analysis from a small number of data sets to a larger batch of data sets in a methodical manner. For example, the algorithms will be capable of recognizing and correcting systematic/repetitious errors that are corrected by the user.
• Simple Interfaces: The user interface will be intuitive, yet powerful enough to express sophisticated problems. The interfaces themselves will be modular and extensible.
In summary, we are really interested in fostering a systematic cross-disciplinary research and development effort whose end product is the FARSIGHT toolkit. This toolkit is expected to be continuously refined in a cross-disciplinary environment.
