Patent Application
20070177782
The invention disclosed herein relates to a method and apparatus for setting the detection threshold used for processing medical images.
Over the last several years, computer aided detection systems have been developed that provide a “second reading” of a medical image such as a mammogram or a chest x-ray. Such systems take a mammogram or x-ray in digital form, process it to locate abnormalities, rank the abnormalities by a scoring algorithm and then display their results in the form of an annotated map that shows the locations of the suspected abnormalities and some measure of their ranking.
The ranking is typically a confidence level that the suspected abnormality is indeed truly an abnormality or a true positive. Alternatively, the ranking could be an estimate of the severity of the suspected abnormality. The ranking can be expressed in numerous ways. For example, it can be expressed numerically alongside a marker indicating the location of the suspected abnormality, or it can be expressed using different size markers for different probability or confidence levels, or different colors, or different shades of the same color. The use of such markers is described, for example, in S. P. Wang's U.S. Pat. No. 6,266,435 for “Computer-Aided Diagnosis Method and System” which is incorporated herein by reference. Examples of commercial use of such markers are found in the assignee's Mammochecker® mammography system and Imagechecker® MSCT system.
As described in the '435 patent, some of the suspected abnormalities will turn out to be of no concern. In other words, they are false positives. Indeed, in any detection system, there is a trade off between sensitivity, or the fraction of true positives detected, and specificity or the fraction of false positives detected. This sensitivity/specificity trade off is often depicted in the detection system's receiver operating characteristic (ROC) curve such as that shown in
In the medical arts, the trade-off between sensitivity and specificity that is represented by the ROC curve is always a concern. If the detection system is not sensitive enough, it may report too few true positives (i.e., more false negatives) which typically represent missed opportunities to detect some sort of problem that may well be life-threatening. On the other hand, if the detection system is not specific enough, it may report too many false positives which typically will result in the performance of additional medical procedures to establish the true nature of the false positive and, in many cases, considerable emotional stress on the part of the patient. Faced with this trade-off, the medical practitioner is usually forced to set the threshold of his/her detection system by trial-and-error at some value that assures the detection of significant numbers of true positives at the cost of some false positives.
This problem is especially acute in the computer-aided detection (CAD) of lung modules in computer tomography (CT) images. Because large numbers of false positives are present in many CT lung images, it is desirable to permit the medical practitioner to vary the detection threshold. Usually, this is done by entering into the computer system a numerical value. The computer system will then report or mark those inputs that have a score that is in excess of the threshold and will ignore all other events. The number of false positives, however, will vary from one lung image to another and the need to suppress false positives will vary accordingly. Thus, the threshold value required to suppress excess false positives will also vary. In addition, the numerical ranges used in scoring the abnormalities are based on arbitrary scales and will vary from detection system to detection system. The result is that the medical practitioner is often forced into some trial and error process for setting the threshold or is required to consult some other source to determine a suitable threshold.
In accordance with the present invention, a more intuitive user interface is provided which allows for selection of the abnormality level on the basis of graphical depictions of the abnormalities themselves. In particular, in one embodiment the user interface displays realistic images of actual abnormalities in a series ranging from least severe to most severe and the medical practitioner can select the abnormality level by mouse-clicking on one of the images. The user interface will then display the annotated map of suspected abnormalities with annotations only on those suspected abnormalities that are at least as severe as the selected abnormality.
In alternative embodiments, the user interface displays stylized images of the abnormalities plotted by the display using mathematical functions. In yet another alternative, the images are hand-drawn images.
These and other objects, features and advantages will be more readily apparent from the following Detailed Description in which:
The processing that is performed includes the processing that creates a visual image from the x-ray data obtained by the scanner. Advantageously, the processing may also include processing that combines the two-dimensional image data on each slice to form a perspective view of a three-dimensional anatomical structure. In addition, the processing may also include analysis of the digital images to detect features that may be of interest to a medical practitioner, e.g., features indicative of cancer and the like. One such prior art system is the assignee's ImageChecker® CT system. Certain details of this system are described in U.S. Pat. No. 6,925,200 which is incorporated herein by reference.
As indicated above, in prior art detection systems the medical practitioner sets the threshold for marking abnormalities in the CT lung images by entering a numerical value at the computer keyboard. The CAD system will then mark as abnormal only those features in the images that have a score that is greater than the threshold.
In the present invention, there is no need to enter a numerical value. Rather, a set of images are displayed on display 220 representative of the features that might be observed in a CT lung image. The images are displayed in order ranging from least severe to most severe and they are related by the CAD system to specific numerical scores. As a result, the medical practitioner can set a threshold visually by looking at the image features that concern him and then clicking on one of the images. The CAD system records which image was clicked and obtains the numerical score associated with that image. It then marks as abnormal only those features in the image that have a score that is greater than the numerical score and displays those features with the appropriate markings.
In one embodiment of the invention, the images displayed on display 220 are realistic images of actual abnormalities in a server ranging from least severe to most severe. Three sets of such images of lung nodule abnormalities are shown in
Alternatively, the images that are used could be synthetic images such as those shown in
While the invention is particularly useful in the setting of thresholds for CT lung images, it may also be practiced in setting other thresholds. For example, it may be used in setting the threshold for a CAD system used to mark abnormalities on mammograms. Again, the images that are used may be images of actual abnormalities or synthetic images representative of such abnormalities. Alternatively, they may be hand drawn sketches of typical abnormalities. One such set of sketches for detecting abnormalities in mammograms is shown in
