Catálogo de publicaciones - libros

An improved image similarity method is introduced to recognize breast cancer, and it is incorporated into a computer-aided breast cancer detection system through Bayes Theorem. Radiologists can use the differences between the left and right breasts, or asymmetry, in mammograms to help detect certain malignant breast cancers. Image similarity is used to determine asymmetry using a contextual and then a spatial comparison. The mammograms are filtered to find the most contextually significant points, and then the resulting point set is analyzed for spatial similarity. We develop the analysis through a combination of modeling and supervised learning of model parameters. This process correctly classifies mammograms 84% of the time, and significantly improves the accuracy of a computer-aided breast cancer detection system by 71%.

The purpose of this study was to investigate the usefulness of multiple-view mammograms in the computerized scheme for identifying histological classifications. Our database consisted of mediolateral oblique (MLO) and craniocaudal (CC) magnification mammograms obtained from 77 patients, which included 14 invasive carcinomas, 17 noninvasive carcinomas of comedo type, 17 noninvasive carcinomas of noncomedo type, 14 mastopathies, and 15 fibroadenomas. Five features on clustered microcalcifications were determined from each of MLO and CC images by taking into account image features that experienced radiologists commonly use to identify histological classifications. Modified Bayes discriminant function (MBDF) was employed for distinguishing between histological classifications. For the input of MBDF, we used five or ten features obtained from MLO and/or CC images. With ten features, the classification accuracies for each histological classification ranged from 70.6% to 93.3%. This result was higher than that obtained with only five features either from MLO or CC images.

In this paper we investigate the effect of reclassification of CAD findings using correspondences in MLO and CC views, with the aim of reducing false positives and inconsistencies. We use a method to link regions identified as suspicious in both projections and add a two-view classifier to an existing CAD scheme. The input of this two-view classifier was a feature vector containing the likelihood of malignancy of the region, the likelihood of malignancy of the corresponding region in the other view, and a number of features that describe the resemblance between the both regions. Using FROC analysis we show that detection results improve when using two-view information.

Standard Mammogram Form (SMF), is a standardized, quantitative representation of a breast x-ray that can be easily estimated. From SMF it is possible to compute the volume of non-fat tissue and measures of breast density, both of which are of significant interest in determining breast cancer risk. Previous theoretical analysis of SMF suggested that a complete and substantial set of calibration data (such as mAs and kVp) would be needed to generate realistic breast composition measures, which is problematical since there have been many interesting trials that have retrospectively collected images with no calibration data. In this paper, we show how implementations of SMF include self-compensation mechanisms, so that SMF can be applied retrospectively to data for which calibration parameters are not (or only partially) available. To illustrate our findings, the current implementation of SMF (version 2.2 β ) was run over 4,028 digitized film-screen mammograms taken from 6 sites during the years 1988-2002, both with and without using the known calibration data. Results show that the SMF implementation running with no calibration data generates results which display a strong relationship with those obtained using a complete set of calibration data. More importantly, they bear a close relationship to an expert’s visual assessment of breast composition using established techniques.

In order that accurate measurements of volumetric breast density may be made, a model of the scattered radiation present within an image is required: such a model is presented here. The model has the advantageous property of utilising a model of photon scattering, allowing cross sections to be calculated, and thus allowing scatter to be modelled for any object. An analysis is presented which uses the model to quantify the effect of varying small angle scattering properties of breast tissues; and the effect of the height within the breast at which tissues are present. Since the details of the anatomical structure of the breast under measurement are unknown, their precise effect on scatter cannot be calculated, but this model is used here to establish error bounds on the scatter estimate, which is a significant contribution to the error in breast density measurement.

In the Netherlands a number of (screening) trials with digital mammography equipment have been started since 1999. In this paper results from the weekly QC procedure are given. It seems that the homogeneity test as described in the addendum to the European protocol is able to detect detector problems and flat field calibration problems. However, visual inspection remains necessary. For the CR system in the trials the homogeneity test did not find many problems. Either the homogeneity test is not effective and other tests might be more appropriate or this CR system does not have relevant image quality variations and therefore might not require weekly quality control.

European Guidelines for quality control in digital mammography specify minimum and achievable standards of image quality in terms of threshold contrast, based on readings of images of the CDMAM test object by human observers. However this is time-consuming and has large inter-observer error. To overcome these problems a software program (CDCOM) is available to automatically read CDMAM images and can be used to predict the threshold contrast for a typical observer. The results of threshold contrast determination by a panel of 3 human observers was compared in this study to predicted human readings for different types of digital mammography system to determine whether this provides a viable method of automated quality control and comparison with existing European Guidelines.

Optimization of acquisition technique factors (target, filter, and kVp) in digital mammography is required for maximization of the image SNR, while minimizing patient dose. The goal of this study is to compare, for each of the major commercially available FFDM systems, the effect of various technique factors on image SNR and radiation dose for a range of breast thickness and tissue types. This phantom study follows the approach of an earlier investigation [1], and includes measurements on recent versions of two of the FFDM systems discussed in that paper, as well as on three FFDM systems not available at that time. The five commercial FFDM systems tested are located at five different university test sites and include all FFDM systems that are currently FDA approved. Performance was assessed using 9 different phantom types (three compressed thicknesses, and three tissue composition types) using all available x-ray target and filter combinations. The figure of merit (FOM) used to compare technique factors is the ratio of the square of the image SNR to the mean glandular dose (MGD). This FOM has been used previously by others in mammographic beam optimization studies [2],[3]. For selected examples, data are presented describing the change in SNR, MGD, and FOM with changing kVp, as well as with changing target and/or filter type. For all nine breast types the target/filter/kVp combination resulting in the highest FOM value is presented. Our results suggest that in general, technique combinations resulting in higher energy beams resulted in higher FOM values, for nearly all breast types.

A digital full-filed mammography system using phase-contrast technique has been developed. The system consists of a dedicated mammography unit, a computed radiography unit with a sampling rate of 43.75 microns, and a photothermographic printer with a printing rate of 25 microns for photothermographic film with the maximum optical density of 4.0. The sharpness of the output image is improved with an edge effect due to phase contrast and magnification. The image noise is reduced by an air-gap method with no bucky. In this paper, the image qualities of the phase-contrast mammography are described for full-filed mammography and spot-compression at 1.5x magnification.

The Multiple Image Radiography (MIR) method is new imaging modality that extends the capability of conventional absorption based radiography by adding the additional contrast mechanisms of x-ray refraction and ultra-small angle scatter. In order to design a clinically based MIR system, the MIR specific x-ray properties in breast tissue must be analyzed to determine which are diagnostically useful. Developing MIR as an imaging modality also requires developing new phantoms that incorporate x-ray refraction and ultra-small angle scatter in addition to traditional x-ray absorption. Three breast cancer specimens were imaged using MIR to demonstrate its MIR specific x-ray properties. An uncompressed anthropomorphic breast phantom with an imbedded low absorption contrast acrylic sphere was imaged to provide a physical model of how the unique properties of MIR can be utilized to improve upon conventional mammography and illustrate how these can be used to design a clinically useful imaging system.