Application of ROC curve to evaluate the of characteristic parameters of diagnosis value microcalcifications on breast cancer
Liang Wen-zha · Journal of Practical Medical Imaging · 2013
Objective To apply ROC curve to investigate the diagnostic value of distribution and morphological characters of microcalcifications for breast cancer.Methods The image data of 354 patients who were found microcalcifications with mammography and proved breast carcinoma(231 cases),benign lesions(123 cases) by pathology,was detected by computer-aided detection system(CAD) with characteristic parameters as follows: the number of unit area,average regional diameter,difference of length,rate of sand-like and linear branching-like calcification.ROC curve was used to analyze the clinical properties and discover their cut off point.Results The cut off point of above parameters were as follows:32/cm2、23.11 mm、1.30 mm、79.3%、50.9%,and the under-ROC curve areas were 0.735、 0.750、0.678、0.850、0.890 respectively.The linear branching-like ratio revealed the highest diagnostic value with specificity 89.0%,sensitivity 70.0%.Conclusions ①The characters of microcalcifications significant variously in diagnostic value,single detection and anylasis of which is inadequate.Accurate diagnosis requires comprehensive application of multiple parameters.②ROC curve can be used to evalue the diagnostic value and get the cut off point of respective characteristic parameter,which means significantly to the development of CAD technology and image diagnosis work.