Reyer Zwiggelaar

Detection of Mammographic Microcalcifications Using a Statistical Model

Breast cancer is the leading cause of early mortality in women [1]. Reseach has shown that radiologists involved in screening mammograms for signs of early breast cancer can be aided by the provision of prompts to direct their attention towards potential abnormalities. In order for prompting to be successful in improving detection performance, the error rates of prompt generation algorithms must be strictly controlled [2]. Almost half of clinically occult breast cancers are due to the presence of microcalcifications [3]. In this paper, a new method is proposed to achieve the automatic detection of microcalcifications. A directional recursive median filtering (DRMF) technique at various scales and orientations is applied to the mammograms to obtain signatures at a pixel level which are characteristic of the local greylevel distribution [2], [4]. We have developed a Principal Component Analysis (PCA) statistical model based on the signatures [2], [4] which can be used for the detection of microcalcifications. A Receiver Operating Characteristic (ROC) study based on pixel classification is provided and the results are compared with approaches published in the literature [5], [6].

Prompting in Mammography: How Good Must Prompt Generators Be?

The UK breast screening programme generates in excess of one and a half million X-ray mammograms per year, all of which must be carefully searched for any signs of disease by experienced radiologists. Signs of early cancer are often small or subtle, and occur only infrequently. The consequences of errors are costly, and in many centres, films are read by two experienced radiologists in an attempt to improve accuracy.

Tracking mammographic structures over time

A method to correspond linear structures in mammographic images is presented. Our approach is based on automatically establishing correspondence between linear structures which appear in images using robust features such as orientation, width and curvature extracted from those structures. The resulting correspondence is used to track linear structures and regions in mammographic images taken at different times. Medical image analysis [1] has been an important research subject in recent years where computer vision techniques have been successfully applied to develop detection and diagnosis systems, enhancement and training tools. The analysis of mammographic images is one of those fields and as such a very challenging one due to the complexity of the images and the subtle nature of the abnormalities. Detection of abnormal structures or architectural distortions in mammographic images can be performed by analysing different images of the same patient. Various approaches have been adopted which bring images into alignment in order to detect differences which are likely to be due to an abnormality. A large number of those methods are based on automatically corresponding extracted landmarks from mammographic images. Those landmarks include breast boundary [15, 20, 9], pectoral muscle [9], salient regions extracted using wavelets [14], iso-intensity contours [11] or steerable filters [20] and crossing points of horizontal and vertical structures [22]. This work presents an approach to the correspondence in mammographic images based on anatomical features which appear as linear structures in the images. The correspondence is used here to track linear structures in mammograms of the same patient over several years. Tracking of linear structures could be used to assess and model the development of architectural changes and abnormal structures. By being able to track regions back in time the available information will help to improve early detection of subtle abnormalities which are initially missed by radiologists. The tracking of objects in image sequences is a well-developed area [21]. However, in general this involves rigid objects (like cars [8]) or objects with a predictable behaviour (like humans [10] or animals [18]). Another difference with the current application is the fact that normally tracking is established using sequences of tens to hundreds of images and not only a few.

Abnormal Masses in Mammograms: Detection Using Scale-Orientation Signatures

We describe a method for labelling image structure based on scale-orientation signatures. These signatures provide a rich and stable description of local structure and can be used as a basis for robust pixel classification. We use a multi-scale directional recursive median filtering technique to obtain local scale-orientation signatures. Our results show that the new method of representation is robust to the presence of both random and structural noise. We demonstrate application to synthetic images containing lines and blob-like features and to mammograms containing abnormal masses. Quantitative results are presented, using both linear and non-linear classification methods.

Detection of the Central Mass of Spiculated Lesions - Signature Normalisation and Model Data Aspects

We describe a method for labelling image structure based on non-linear scale-orientation signatures which can be used as a basis for robust pixel classification. The effect of normalisation of the signatures is discussed as a means to improve classification robustness with respect to grey-level variations. In addition, model data selection and scale normalisation are investigated as a means to improve the robustness of detection with respect to the scale of structures.

Separating background texture and image structure in mammograms

There have been several approaches to the classification of texture in images. Most approaches will take certain local attributes or features into account and base the classification on these measures. In here we demonstrate the use of a statistical approach to separate the structure and texture background present in images. Modelling is based on normal images which only contain a texture background. The resulting model is applied to images which contain abnormal image structures as well as a normal texture background. Especially for mammographic (and other medical application) this can provide useful information which can be used as a pre-processing tool to obtain the structures present in the image and at the same time get a robust classification of the background texture.

Detecting the central mass of a spiculated lesion using scale-orientation signatures

Potential malignancies in mammograms can be detected from subtle abnormalities in radiographic appearance. Radiologists fail to detect a significant proportion of such abnormalities, but it has been shown that their performance would improve if they were prompted with the locations of possible abnormalities [1].

Automatic Classification of Linear Structures in Mammographic Images

Certain kinds of abnormalities in x-ray mammograms are associated with specific anatomical structures – in particular, linear structures. This association can, in principle, be exploited to improve the specificity and sensitivity with which the abnormalities can be detected. We compare annotated and the automatic detection of the scale and orientation associated with linear structure in mammograms. We investigate methods of classifying the detected structures into anatomical classes (spicules, vessel, duct, fibrous tissue etc) from their cross-sectional profiles. Automatic (linear and non-linear) classification results are compared with expert annotations using receiver operating characteristic analysis. We show that useful discrimination between anatomical classes is achieved. Some of this relies on simple attributes such as the width and contrast of the profile, but there is also important information carried by the shape of the profile.

Comparison of Methods for Combining Evidence for Spiculated Lesions

Breast cancer is a leading cause of fatality in women, with approximately 1 in 12 women affected by the disease during their lifetime [1], Mass screening of women using x-ray mammography is currently the most effective method of early detection of the disease; this is essential for successful treatment [2].

Linear and Non-Linear Modelling of Scale-Orientation Signatures

We describe a method for labelling image structure based on scale-orientation signatures. These signatures provide a rich and stable description of local structure and can be used as a basis for robust pixel classification. We demonstrate their application to synthetic images containing lines and blob-like features and to mammograms containing abnormal masses. Quantitative results are presented, using both linear and nonlinear classification methods.