C. E. Metz, "Evaluation of digital mammography by ROC analysis," in Digital Mammography (International Congress Series), K. Doi, Ed. Amsterdam, The Netherlands: Elsevier, 1996, pp. 61--68.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....trained. A ROC curve that was generated with the same dataset that was used to train the classifier is referred to as a consistency ROC curve. A validation ROC curve is obtained when the curve is generated from an independent data set and represents an unbiased estimate of classifier performance [24]. Two typical ROC curves are shown in Fig. 2. The area under a ROC curve, or , is an accepted way of comparing overall classifier performance [6] 7] 20] 21] Two curves may have equal values, as shown in Fig. 2. However, one of the curves will typically be preferred over the other, depending ....

C. E. Metz, "Evaluation of digital mammography by ROC analysis," in Digital Mammography (International Congress Series), K. Doi, Ed. Amsterdam, The Netherlands: Elsevier, 1996, pp. 61--68.

....signatures can be used directly in pixel classification. As an example, the method has been applied to the detection of abnormal masses associated with spiculated lesions. Detection performance results are presented using receiver operator characteristic (ROC) and free response ROC (FROC) curves [2]. The results are compared with other mass detection methods [3 5] and suggestions for possible improvements are made. 2 Directional Recursive Median Filtering The Recursive Median Filter (RMF) is one of a class of filters, known as sieves, that remove image peaks or troughs of less than a ....

.... by p ij = 1 (2 ) n=2 jC i j 1=2 exp ( Gammaffi ij 2 ) 1) The odds ratio of class not class for each pixel j is given by OR ij = p ij P n6=i p nj : 2) Applying (2) to every pixel results in an odds ratio image which can be thresholded at various values to generate ROC FROC curves [2]. 3 Detecting Abnormal Masses in Mammograms We present results for a set of 28 mammograms from the PRISM database, all containing spiculated lesions; these had been annotated by an expert radiologist. A typical example of a mammogram containing a mass like structure associated with a spiculated ....

C.E. Metz. Evaluation of digital mammography by roc analysis. Excerpta Medica, 1119:61--68, 1996.

....measure, based on experimental evidence, that is particularly relevant to data sets in which the proportion of abnormalities is higher than the screening average. We also describe, briefly, how different detection techniques can be compared through the use of receiver operating characteristics (Metz, 1996). The statistical methods on which both approaches are based are discussed briefly in section 3, whilst sections 4 and 5 give in depth descriptions of the methods developed for the detection of the abnormal patterns of linear structures and central masses respectively and the results obtained. The ....

....system can still lead to an improvement in radiologists performance. 2.1. ROC and FROC analysis The performance of both radiologists and computer based detection techniques can be assessed using receiver operating characteristic (ROC) and free response operating characteristic (FROC) curves (Metz, 1996). An ROC curve (e.g. Figure 2) indicates the true positive rate (sensitivity) as a function of the false positive rate (1 Gammaspecificity) When no useful discrimination is achieved the true positive rate is always equal to the false positive rate. As the accuracy increases the ROC curve moves ....

[Article contains additional citation context not shown here]

Metz, C.E. (1996) Evaluation of digital mammography by roc analysis. Excerpta Medica, 1119, 61--68.

....to generate more false positive responses. Using a large threshold gives fewer false positive responses, but may miss more true lesions. Hence variation of thresholds gives di#erent diagnostic accuracy which can be quantified using FROC ( Free response Receiver Operating Characteristic ) analysis [23], where the true positive fraction (TPF) is plotted as a function of the average number of false positives (FP) per image. FROC analysis [23] is well suited for the assessment of computer aided diagnosis of mammograms because it is applicable to situations that involve any number of reported ....

....variation of thresholds gives di#erent diagnostic accuracy which can be quantified using FROC ( Free response Receiver Operating Characteristic ) analysis [23] where the true positive fraction (TPF) is plotted as a function of the average number of false positives (FP) per image. FROC analysis [23] is well suited for the assessment of computer aided diagnosis of mammograms because it is applicable to situations that involve any number of reported locations and any number of actual lesions in each image. If there is a positive detection at a coarser resolution, no feature extraction and ....

C. E. Metz, "Evaluation of digital mammography by ROC analysis," Proceedings of the 3rd International Workshop on Digital Mammography, June 9--12 1996, Chicago,

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC