An improved wavelet-based contralateral subtraction algorithm

Contralateral subtraction is an image processing technique aimed to enhance asymmetries in chest radiographs. Since malignant processes often result in asymmetrical lesions, a computer aided diagnostic system capable of enhancing asymmetries on these images could support the diagnostic procedure. The technique itself consists of registration of the two lung areas, and subtraction of the registered images. Contralateral subtraction is therefore essentially a registration problem, a field that has been widely researched offering various solutions to the problem at hand. One of these solutions is wavelet transformation, a multiresolution method that has a potential not fully unlocked by previous methods. We propose a contralateral subtraction scheme based on an elastic registration algorithm in the wavelet domain, using dual tree complex wavelet transform. By registering only the low frequency components of the images, the resulting transformation will be inherently regular, without relying on conventional regularization methods. We demonstrate the effectiveness of this concept by applying it to a contralateral subtraction problem, verifying the algorithm with synthetic tests as well as chest radiographs from the JSRT public database. These results show an increased sensitivity of nodule detection, indicating that the proposed method is a valid alternative to other known registration methods.