A 3D α-hull approach to juxta-pleural nodule detection in chest Computed Tomography images

ROI analysis and feature extraction provide useful attributes to characterize in a specific manner the content of a given region of interest, allowing the next stage of classification to distinguish between e.g. healthy and pathological patterns. The choice of appropriate features is crucial for the CAD to work optimally, with good diagnostic accuracy.

Characterizing features requires a specific study of many parameters that we can classify in two large groups: those calculated from the gray-value distribution of pixels in the ROI, and those "measuring" the shape of the objects contained in the ROI. Texture is the term used to characterize the tonal or gray-level variations in an image. Texture analysis is the joint study of pixel's geometrical location and intensity in the ROI.

There are two major types of texture approach, the structural and statistical. Structural texture is referred to the research of structures which are repeated, showing general characteristics that are very similar to each other, such as oriented lines, with a (more or less) deterministic rule of displacement. These elements, called primitives, may be extracted with many methods [22] and then identified. Structural approaches to texture analysis aim to discern the textural primitive and to determine the underlying gross structure of the texture.

Properties of the primitives (e.g. area and average intensity) are used as texture features and the analysis is completed either by computing statistics of the primitives (e.g. intensity, area, elongation, and orientation) or by deciphering the placement rule of the elements. The advantage of the structural methods is that they provide a good symbolic description of the image. Statistical textures on the contrary, cannot be described with displacement rules and statistical texture analysis always represents texture indirectly by non-deterministic properties. The spatial distribution of gray values is therefore analyzed by computing local features at each point in the image. This gives the possibility to collect a set of statistics from the distributions of the local features.

Furthermore, statistical methods are classified according to the number of pixels considered in the analysis. Basically there are first order and second order statistical parameters but we can extend the analysis to higher orders. First-order statistical methods refer to the local features that are defined only by one pixel excluding all the statistics due to pixel neighborhood relationships. Texture analysis based solely on the gray-level histogram suffers from the limitation that it provides no information about the relative position of pixels to each other. For example, two completely different images each with a 50% black and 50% white pixels may produce the same gray-level histogram. Therefore we cannot distinguish between them using first order statistical analysis. This problem is solved by second-order statistical methods or higher methods that estimate joint properties of two or more pixel values occurring at specific locations relative to each other. There are also other texture feature analysis methods such as model- based methods [23]. Here an attempt is made to represent texture in an image using sophisticated mathematical models (such as fractal or stochastic).