Computed Tomography (CT) devices irradiate a (human) body with controlled amounts of X-ray to produce an image in which each substance (lung, tissue, vessels, etc.) can be identified unequivocally. Commonly, CT devices also capture areas that do not belong to the human body. Such areas are referred to as air, or air pixels, and may contain imaging artifacts. The air pixels are irrelevant for the medical diagnostic and provoke an important degradation in coding efficiency. Work on CT images has shown that these two drawbacks can be avoided by removing the air pixels. This paper presents a filter that removes air pixels in the wavelet domain with an accuracy that can be selectively adjusted and that requires negligible computational resources. The main insight behind the proposed filter is a theoretical analysis that models air pixels in the wavelet domain. Experimental results in the framework of JPEG2000 indicate a coding gain of more than 3 bpp.