Keeping Informed: Automatic Processing of Residual Functional Capacity Form Images

Abstract

The U.S. Social Security Administration (SSA) must process millions of claims for disability benefits each year. The statutory definition of disability looks at the expectation of gainful employment in the presence of severe impairment. A key component of this is a person's functional ability -- both physical and mental -- to work. SSA summarizes a claimant's functional ability in both checkboxes and free text within the Residual Functional Capacity (RFC) Assessment forms. Until recently, these were paper forms that were stored as images in SSA's databases. In order to access the data in these forms for comparison to other measures of function and to assist in SSA's business practices, it is necessary to have a way to automatically extract the data captured in these forms. However, there is a lot of variability across RFC forms that available software could not capture. Therefore, we designed a new system that relied on basic computer vision processes, novel form templates based on the checkbox grid, or row-column, structure, and optical character recognition to convert the images to text in order to extract all data related to function. The robustness of this method comes from the templates and the use of the grid structure rather than checkbox location.