Computed-Aided Daignosis For ICU Images

Chest radiographs are useful in ICU (Intensive Care Unit) for evaluation significant or unexpected pathological changes of the patient and estimating the accurate position of inserted medical devices. During a treatment period in the ICU, a number of images are in common obtained periodically; however, these ICU images could be inconsistent due to the differences between exposure settings, patient and apparatus positioning and the change of environmental condition. In some cases, the change of some anatomical structures in the periodical images is very subtle but highly significant, which could be overlooked by physicians. This will result in an operational difficulty in comparatively interpreting the change of those temporal images. To circumvent this problem, we investigate a novel consistent rendering scheme in diagnosis of periodic Chest X-ray images, in particular for ICU images. This paper proposes a ROI (Region of Interest) based LUT (Look-up table) mapping algorithm for processing ICU raw images. The structure of this presentation is organized in five sections: background segmentation, automated ROI selection, ROI-based LUT construction, noise reduction LUT mapping and image enhancement. Experimental results demonstrate that our novel image-rendering technique reduces the variability in the image appearance caused by the differences in patient and apparatus positioning and image acquisition parameters(from average rate 3.52 to 4.32). It can potentially improve the overall workflow and patient management.