Catálogo de publicaciones - libros

The objective of the study was to assess the potential of a CAD device with computer aided classification capabilities to reduce interventional procedures for BI-RADS category 4A lesions. 113 such lesions (17 masses, 96 clusters), forwarded for biopsy (103 benign) were analyzed retrospectively by a CAD device that generated descriptors. The device extracted quantitative features characterizing the lesions by shape, margins, size and distribution. Descriptors taken from the BI-RADS lexicon for the appearance of the lesion were generated based on the values of the quantitative features. A paradigm based on the computer generated descriptors was developed to assist in assigning a level of suspicion. The paradigm deemed malignant, all 10 malignant cases of the study (100% sensitivity) and correctly classified 38 of the 103 benign lesions. The CAD-generated descriptors, thus, eliminated 36.9% of unnecessary biopsies without decreasing the sensitivity.

Computer aided detection systems for mammography typically use standard classification algorithms from machine learning for detecting lesions. However, these general purpose learning algorithms make implicit assumptions that are commonly violated in CAD problems. We propose a new ensemble algorithm that explicitly accounts for the small fraction of outlier images which tend to produce a large number of false positives. A bootstrapping procedure is used to ensure that the candidates from these outlier images do not skew the statistical properties of the training samples. Experimental studies on the detection of clusters of micro-calcifications indicate that the proposed method significantly outperforms a state-of-the-art general purpose method for designing classifiers (SVM), in terms of FROC curves on a hold out test set.

We performed a reading session to examine the validity of computer- aided detection for mammograms using a 3-megapixel liquid crystal display (LCD). Digital mammograms of 225 patients (ROLAD M-IV and FCR9000, 100 (/pixel), were divided into 3 data sets (each set consisting of 75 patients, including 30 with pathologically proven breast cancer) for this reading session. Fifteen physicians interpreted these three data set using 3 different imaging modalities; hard copy, LCD without computer-aided detection (CAD), and LCD with CAD. Then they categorized the images into 4 ranks according to the confidential levels of cancer. Sensitivity and specificity were calculated individually for each of the 3 different modalities, and then ROC analysis was performed. The sensitivity, specificity, and Az values showed no significant differences between LCD with out CAD and hard copy. Also, no significant differences were found between LCD with CAD and the other modalities for these 3 values. The results of this study indicate that it is reasonable to use a 3-megapixel LCD for interpretation of digital mammograms instead of conventional hard copy. Nevertheless, because the usefulness of the CAD system has not been fully ascertained, further studies are required.

Computer-aided detection (CAD) systems place prompts in digital images to attract readers’ attention to potential malignancies. A reader must then decide whether or not prompted regions correspond to genuine abnormalities and has the option of disregarding falsely prompted regions. In this paper we investigate different readers’ performance with CAD in the context of breast screening. In a retrospective study, eight consultant radiologists each read over 1000 screening mammograms comprising normal cases, screen detected cancer cases and cases that were detected as cancers subsequently. We present their results in terms of cancer detection and recall rates, and relate this to their previous experience of film reading. Our results show that the detection of cancers did not differ significantly between readers, although more experienced film readers were less likely to recommend that normal cases should be recalled.

We evaluated the impact of integration of CAD (Computer-Aided Detection) system and human observers in digital mammography. We compared the diagnostic efficacy of non-informed observers and informed observers regarding the CAD system’s ability (average FP (false positive) per four images and sensitivity of microcalcifications and mass) to detect cancer. With the informed-group, we previously informed them of the accuracy of CAD. In each group, observers recorded the diagnosis before utilizing the CAD system and after utilizing the CAD system according to BI-RADs category and to six additional categories associated with diagnostic confidence. Regarding diagnostic accuracy, with the informed group, sensitivity and NPV were improved without an increase in FP. On the other hand, the diagnostic accuracy of human observers was influenced by prior notification of CAD’s accuracy and by CAD’s performance in cancer detection itself.

Many rural areas do not have reliable or adequate access to breast cancer evaluation facilities and care. With the advent of digital mammo-graphy it is possible to send high quality mammographic images across teleradiology/- telemedicine networks for interpretation at certified mammo-graphy centers. We have a statewide telemedicine network upon which telemammography is conducted with a number of very rural locations. Strict turn-around times for interpretation are guaranteed in contractual agree-ments. We are also testing the use of ultra-rapid pathology clinics for women with positive mammograms and real-time consultation with oncologists to reduce the time it takes for rural women to receive care.

Currently, the majority of lumpectomy and excisional biopsy procedures are performed using the wire localization (WL) technique; however, this technique suffers from several drawbacks including inaccuracy in placement of the wire, possible displacement of the wire prior to surgery, and ambiguity of the lesion’s location along the wire. We propose dual modality surgical guidance (DMSG) as a means to overcome many of the problems associated with WL. The approach uses a dual modality (digital mammography and breast scintigraphy) breast imaging system developed in our lab to place a small radioactive marker (a radiomarker), directly into the lesion. Here we present the results of measurements of the localization and injection accuracy of our system. The localization accuracy, evaluated by determining the difference between the known and measured inter-well separations, were within 0.76 mm (standard deviation of 0.46 mm) of the true distances for x-ray imaging and within 0.66 mm (standard deviation of 0.43) for gamma imaging. Our maximum error in injection accuracy in any of the three Cartesian coordinates was 1.8 mm. On average, the errors were 0.6, 0.4, and 0.9 mm for x, y, and z respectively. The results of these phantom tests provide encouragement that our upright digital mammography unit can accurately a) locate a lesion in three dimensions, b) inject a radiomarker into the lesion, and c) assess the offset between the lesion and radiomarker centers.

Two-view mammography is known to be more effective than one-view in increasing breast cancer detection and reducing recall rates. In addition, there is evidence that computer aided detection (CAD) systems are able to prompt malignant abnormalities that have been overlooked by a human reader. Using data from the UK NHS Breast Screening Programme (NHSBSP) we compared double reading with single reading using a CAD system, to assess the relationship between CAD and number of views in terms of the sensitivity of the screening regime to cancer detection and the recall rate of normal cases. CAD appeared to contribute to an increased cancer detection rate with single-view mammography without significantly increasing the recall rate. For two-view mammography, there was no significant change in sensitivity using CAD but a significantly higher recall rate. However, single-view mammography was used in incident rounds in which previous mammograms were available whereas two-view mammography was used in the prevalent round where no previous mammograms were available.

We have analysed data from a subgroup of thirty-nine women who had previously gained more than 10kg in adult life, and who were amongst those recruited from a family history clinic to a study examining the effects of diet and exercise on breast cancer risk. At entry to the study and after 12 months they underwent a series of investigations, including mammography during which markers were attached to the compression plate to allow accurate measurement of breast thickness. A calibrated stepwedge was placed adjacent to the breast to enable quantitative analysis. The proportions of glandular and fatty tissue were calculated at each pixel from the stepwedge and thickness data and from these, the percentage gland in the breast was computed, both by area and by volume. Statistical analysis showed that the volume of glandular tissue was not related to breast size. Over the 12 month period, the majority of the women lost weight, while some gained weight. It was found that weight change was correlated with change in the volume of fat in the breasts, with those women who lost the largest amount of weight showing the greatest reduction in volume. There was little change in volume of glandular tissue for any of the women. Percentage gland is often used as an indication of risk of developing breast cancer. These results suggest that measures of percentage of gland (e.g. Boyd groups) may be dominated by excess breast fat in overweight women.

Tomosynthesis provides a major advance in image quality compared to conventional projection mammography by effectively eliminating the effects of superimposed tissue on anatomical structures of interest. Early tomosynthesis systems focused primarily on feasibility assessment by providing 3-dimensional images to determine performance advantages. However, tomosynthesis image quality depends strongly on three key parameters: 1) detector performance at low dose, 2) angular range and number of projections acquired in the tomosynthesis scan, and 3) reconstruction algorithm processing characteristics used to create slice images from the measured projections. In this work, a new GE mammo-graphy tomosynthesis research system was developed that incorporates key improvements in each of these three areas compared to an early feasibility prototype system in use at Massachusetts General Hospital from 2000 to 2004. The performance gains that can be achieved by these enhancements are cha-racterized, and clinical images acquired with the system at the University of Michigan Cancer and Geriatrics Center are presented. The advanced research system also provides the ability to acquire mechanically co-registered x-ray tomosynthesis and ultrasound images of the breast, and initial dual modality images are also presented.