Multiresolution autoregressive filtering for pneumonia detection in medical images

Digital signal processing provides a variety of algorithmic tools, which are useful in modern medical imaging applications. A novel processing method based upon multiresolution autoregressive filters is proposed for automated detection of child pneumonia in X-ray images. Wavelet functions are utilized in order to construct a multiresolution whitening filter which retains information of the texture of the infected lung as well as structural information regarding the relative position of the infected area in the chest. Structural information is incorporated into the autocorrelation function at lower resolution levels whereas texture information is incorporated at finer resolution levels. Combining the outputs of the whitening filters through scales provides for robust and efficient detection of regions with child pneumonia.