Lung segmentation: easier and faster thanks to new algorithms
Lung segmentation: easier and faster thanks to new algorithms
Interview with Dr. Anja Braune (Postdoctoral Research Fellow) and Professor Marcelo Gama de Abreu (Director of Research), Department of Anesthesiology and Intensive Care Therapy - Pulmonary Engineering Group, University Hospital Carl Gustav Carus Dresden
A look inside the lungs is a time-consuming process. To identify the boundaries of the respiratory organ from surrounding other organs, tissues, and structures requires between 200 and 500 computed tomographic images and subsequent manual markings – an elaborate process that can take up to six hours. An optimized computer program is now able to do this in only a few seconds.
Dr. Anja Braune
In this MEDICA-tradefair.com interview, Dr. Anja Braune and Professor Marcelo Gama de Abreu talk about the optimization of a computer program for lung segmentation, explain the advantages of this technology and reveal what else might and should be possible in diagnostic imaging in the future.
You are working on the DICOM Analyzer program. How can it be used for lung imaging?
Dr. Anja Braune: We use different imaging techniques to obtain structural and functional information about the lungs and to quantify the ventilation, perfusion, and inflammation of the individual lung segments. To do this, we use computed tomography and positron emission tomography, among other methods. It is necessary to segment the lung, meaning to delineate it since these images also capture surrounding organs and structures such as the heart, rib cage, and diaphragm. The DICOM Analyzer tool helps us by facilitating the automated segmentation of the lungs.
Prof. Marcelo Gama de Abreu
How exactly does the program delineate the lungs?
Braune: First, we capture the thorax or chest using computed tomography. This provides us with three-dimensional data sets from Hounsfield units, which are largely determined by the density of the recorded volumes. To be able to analyze just the lung data, these 3D data sets are first outlined against other areas based on lung-specific Hounsfield units with the help of the DICOM analyzer program. This delineation via Hounsfield units is easy to do if the lung is well-ventilated. This becomes more difficult in the case of lung damage because collapsed areas exhibit similar Hounsfield units to that of their surroundings.
What are the benefits resulting from this for both physician and patient?
Prof. Marcelo Gama de Abreu: We normally have to manually segment lung structures slice by slice. That means an experienced scientist has to manually delineate the pulmonary tissue from the surrounding tissue on the monitor. This can take a long time. We can automate this process by using software. This allows us to speed up the process and enables a much faster analysis of the images.
This might be of interest to patients in respiratory failure where the goal is an objective characterization of lung ventilation. This technique might also be important for patients who suffer pulmonary contusion caused by chest trauma.
left: CT image of the thorax without segmentation; mi: Result of lung segmentation (red) after evaluation of Hounsfield units within the thorax; right: Result of lung segmentation including anatomical properties of the lung such as limitation by heart (blue) and ribs (green).
But this first requires the optimization of the program. What is the biggest challenge in this case?
Gama de Abreu: What works well in normal and hyperventilated areas is much more difficult in a collapsed lung because it exhibits similar Hounsfield units to those of the surrounding tissue. The delineation of the lungs solely based on Hounsfield units is not an option in the case of damaged lungs. That's why algorithms must be implemented in the software, which can make the delineation by using anatomical information.
Braune: So far, the pathophysiological, collapsed areas could only be segmented manually by experts in a very time-consuming process. In the future, the goal is to delineate them automatically using the DICOM Analyzer program. The applicable algorithms subsequently factor in typical anatomical characteristics, such as the lung boundaries consisting of the rib cage or diaphragm.
You obtained the support from your colleagues at the Brandenburg University of Technology (BTU) for this programming. What exactly does this collaboration look like?
Braune: We have asked the Institute of Medical Informatics at the Brandenburg University of Technology Cottbus-Senftenberg to assist us in reworking and optimizing the DICOM Analyzer program. Thanks to their extensive knowledge in medical informatics, the associates and students led by Professor Bönninger can help us to substantially improve the existing program, as well as implement additional algorithms.
Test run of the optimized computer program DICOM-Analyser (from right): Dr. Anja Braune, Prof. Ingrid Bönninger, Nico Gerhardt, Igor Nesterow, Tobias Steinmetzer.
What happens next? What steps are still necessary for this technique to be implemented in practice?
Braune: The DICOM Analyzer program is initially intended for experimental research use only. For the time being, it is not slated for an application in clinical practice. Additional program improvements are designed to apply the program to a variety of lung injuries and computed tomography images at different resolutions.
Finally, what is your perspective on the future of diagnostic imaging: What could or should be possible in this area?
Gama de Abreu: We need techniques that don't use ionizing radiation. And we need methods that require relatively small devices, so they can be performed right at the patient's bedside. Patients in intensive care units come to mind. Plus we also need devices that can be used by others apart from specialists. That's why the devices should not only be smaller, but also easier to use.
The interview was conducted by Elena Blume and translated from German by Elena O'Meara. MEDICA-tradefair.com
You can find exhibitors and products from the fields of eHealth and Imaging methods at MEDICA 2018: