Hello, I'm Shalom Buchbinder, from the Montefiore Medical Center in the Bronx, and I'd like to acknowledge my international colleagues from the Jerusalem College of Technology, and Hadassah Hebrew University Medical Center, that participated in this research. Our research is aimed at improving the ability to evaluate micro-calcifications which are found on mammography. If a cluster of calcifications is found on mammography, and are ... the physician does not have a good feel as to whether it is malignant or benign, potentially a computerized system can be used to diagnose the possibility or the probability of the cluster being malignant. We initially looked at approximately 50 different features of clusters of micro-calcifications, and we determined that there were eight features that were very, very important in differentiating benign from malignant clusters of micro-calcifications.

Initially what we did, was, we um, we had the computer analyze 260 cases of micro-calcification. And it determined values; above a certain threshold, something was assumed to be malignant, below a certain threshold, something was assumed to be benign, and there was a very tight indeterminate range. And, if a cluster of calcification met criteria, it was given a point for malignancy, and, if it met benign criteria, it was given no points for malignancy, or a zero.

We sum the numbers of each one of these features, and we analyzed the summation, and we were able to identify a positive predictive value of about 67 percent, and a true positive fraction over a false positive fraction, a so called ROC curve, at an area of about 0.81. Then we analyzed the clusters of calcification, the different parameters, and we said- not all parameters should be equally weighted.

Not all clusters of calcifications have features that are of equal importance. We know, for example, that a calcification, or a cluster of calcification, that is made up of very, very fine calcifications that are a circle, are very rarely, if ever, malignant. And we know, as a calcification gets to be more and more irregular, it is more likely that it would be malignant. So we said that not all features were equally important, and we actually, again, re-analyzed the 260 cases, and we looked at each one of these features to determine how important was it ultimately in differentiating benign from malignant clusters of calcification?

Then, what we did was, we looked at each one of these features, saying how strongly was it malignant? Okay, and that was given a value. We multiplied by its relative weight in importance of differentiating benign from malignant calcifications, and each one of these points then received a value. And we reanalyzed that data, and we were able to improve the positive predicative value, which is; how often when we recommend something to be biopsied is it indeed going to be malignant? We improved the positive predictive value from a very good 67 percent to a 79 percent. And we improved the accuracy rate from 69 percent to an 83 percent. And again, we kept the sensitivity at 98 percent. So what this system is able to do, in my belief, is that we will be able to decrease the amount of unnecessary biopsies as a first result of this type of analysis. Currently, when a recommendation is made in the community, it has about a ten to 30 percent positive predictive value. So, ten out of a hundred clusters will indeed be malignant, or 30 out of a hundred clusters will be malignant. What we're able to do is raise that number to- of those that we want to be biopsy- we can get up to about 79 percent will be malignant. So we will be able to obviate many, many patients from receiving unnecessary biopsies. Therefore, obviously, you'll alleviate the anxiety, the uncertainty, when you're told that you may have an abnormality. Certainly, financial considerations will play into this, in that the cost for the unnecessary biopsies will be reduced, and we're very, very hopeful that this type of system, in conjunction with the full field digital mammography, will significantly improve mammography's ability to correctly identify abnormalities, and also to correctly identify structures that need not be biopsied.

Back to main page