Keywords: cardiothoracic ratio, chest anatomy segmentation, chest X-ray
TL;DR: Starting from Chest X-ray images, we investigated cardiothoracic ratio computation methods robust to lung clipping or varying patient orientation.
Abstract: The cardiothoracic ratio (CTR) plays an important role in early detection of cardiac enlargement related diseases in chest X-ray (CXR) examinations. Since its measurement would be time-consuming, its evaluation in clinical practice is done by a visual assessment: it is highly subjective and its robustness is undermined by some acquisition issues such as lung clipping or patient orientation variation. No work addressing the problem of clipped lungs in the CTR estimation has been found in the literature. For these reasons, aiming for a robust method, we firstly proposed a segmentation-based approach for automatic measurement of the CTR (based only on the lung segmentation mask) able to handle clipped anatomy cases. Secondly, the proposed method was validated on a large dataset allowing us to corroborate earlier research results with manual CTR computation in which the mean CTR increases with the age of the patients and there is a noticeable difference between men and women's CTR. Lastly, a new rotational invariant metric was proposed, showing it to be more robust to different patient orientations.