A New CNN Occlusion Masking Method for IRT Imaging in Neurosurgery

Abstract

In this paper, a new occlusion masking method based on Cellular Nonlinear Networks (CNN) for a 2-dimensional (2D) InfraRed Thermography (IRT) brain image is proposed. IRT imaging is one of the noteworthy technique in the medical analysis due to capturing those activities which cannot be seen with unarmed eyes. In neurosurgery, the image occlusion can be classified roughly into two fashions namely, surgery tools and reflection of surgery tools (or cover by the ice-cold saline solution) over the brain cortex surface. In the proposed method, firstly, a CNN texture segmentation is carried out to separate the image textures by the local ratio of the black and white pixels. Thereafter, a CNN dilation is employed to increase the black and white pixels separation. Next, a CNN remove-small-object and filling-holes are exerted for revamping the texture segmentation. Afterwards, the obtained mask is applied to remove the occluded area in the image. To overcome the issues related to reconstruction after masking the image, a solution is given by a 5×5 overlapping sliding window with respect to a reference image, in order to substitute the masked area (pixel values) in the target image. Our proposed method is evaluated on real IRT brain cortex images and compared with other methods. The proposed method shows promising results for IRT brain images in comparison to those other mentioned methods.