Characterizing non-linear dependencies among pairs of clinical variables and imaging data.

Advances in computer-aided diagnosis (CAD) systems have shown the benefits of using computer-based techniques to obtain quantitative image measurements of the extent of a particular disease. Such measurements provide more accurate information that can be used to better study the associations between anatomical changes and clinical findings. Unfortunately, even with the use of quantitative image features, the correlations between anatomical changes and clinical findings are often not apparent and definite conclusions are difficult to reach. This paper uses nonparametric exploration techniques to demonstrate that even when the associations between two-variables seems weak, advanced properties of the associations can be studied and used to better understand the relationships between individual measurements. This paper uses quantitative imaging findings and clinical measurements of 85 patients with pulmonary fibrosis to demonstrate the advantages of non-linear dependency analysis. Results show that even when the correlation coefficients between imaging and clinical findings seem small, statistical measurements such as the maximum asymmetry score (MAS) and maximum edge value (MEV) can be used to better understand the hidden associations between the variables.