Improved Lesion Detection by Level-Dependent Spatial Summation

A novel signal postprocessing method has been developed for application in lesion detection. A signal distribution, directly related to the 2-dimensional envelope distribution comprising a B-mode image, is convolved with a block shaped kernel of which both base area (i.e., window size) and height are modulated by the local signal level. A theoretical analysis shows that the output includes information regarding the local mean brightness in the original image. This information can be displayed as a gray scale image. The predicted behavior agrees with experimental observations. On the basis of twenty simulated lesion images, it is demonstrated that the present algorithm is superior to the conventional L₂ -mean filter in lesion detection problems, whereby the performance is quantified by the area A_ROC under the Receiver Operating Characteristic. An improvement in the median value of A_ROC of the order of 0.1 was observed. Two parameters are used for adjustment. The total number of summations involved in the processing of a frame is comparable to the number required by the arithmetic mean filter, allowing real-time implementation.