Simulation Algorithms for Spatial Pattern Recognition

DESCRIPTION (provided by applicant): This SBIR project is developing methods and software for the specification, construction and simulation of neutral spatial models, and for applying these neutral models within the framework of probabilistic pattern recognition. Results will allow epidemiologists, environmental scientists and image analysts across a broad range of commercial disciplines to more accurately identify patterns in spatial data by removing the bias towards false positives that is caused by unrealistic null hypotheses such as "complete spatial randomness" (CSR). This project will accomplish 5 aims: 1. Conduct a requirements analysis to specify the neutral models and functionality to incorporate in the software. 2. Develop and test a software prototype to evaluate feasibility of the proposed models. 3. Propose a topology of neutral models and develop strategies to generate them and to conduct sensitivity analysis for investigating the impact of implicit assumptions (i.e. spatial autocorrelation or non-uniform risk) and number of realizations on test results. 4. Incorporate the neutral models in the first commercially established software package that allows for user-specified alternate hypothesis in spatial statistical tests. 5. Apply the software and methods to demonstrate the approach and its unique benefits for exposure and health risk assessment. Feasibility of this project was demonstrated in the Phase I. This Phase II project will accomplish aims three through five. These technologic, scientific and commercial innovations will revolutionize our ability to identify, document and assess the probability of spatial patterns relative to neutral models that incorporate realistic local, spatial and multivariate dependencies. The neutral models and methods in this proposal make possible, for the first time ever, evaluation of the sensitivity of the results of cluster or boundary analyses to specification of the null hypothesis.