Preprocessing of Heteroscedastic Medical Images

Tissue intensity distributions in medical images can have varying degrees of statistical dispersion, which is referred to as heteroscedasticity. This can influence image contrast and gradients, but can also negatively affect the performance of general-purpose distance metrics. Numerous methods to preprocess heteroscedastic images have already been proposed, though most are application-specific and rely on either manual input or certain heuristics. We therefore propose a more general and data-driven approach that relies on the notion of intensity variance around each specific intensity value, simply referred to as intensityspecific variances. First, we introduce a method for estimating these variances from an image (or a collection of images) directly, which is followed by an illustration of how they can be used to define intensity-specific distance measures. Next, we evaluate the proposed concepts through various applications using both homoand heteroscedastic CT and MR images. Finally, we present results from both qualitative and quantitative analyses that confirm the working of the proposed approaches, and support the presented concepts as valid and effective tools for (pre)processing heteroscedastic medical images.