Lung CT Segmentation to Identify Consolidations and Ground Glass Areas for Quantitative Assesment of SARS-CoV Pneumonia

Since the outbreak of the novel coronavirus disease pandemic, a simple and effective method useful to obtain a proper segmentation of the typical pathological lesions of this infection in chest CT scan images appears fundamental. Thus, our aim is to offer a starter pack of knowledge in order to let people perform an easy and efficient segmentation of these lung pathological findings. Hopefully, this will help the scientific community to develop dedicated radiomics and artificial intelligence studies that could improve the diagnostic approach of these patients.

From the DICOM browser of 3D Slicer, proceed to select the study you want to segment, and click Load. You can review the selected case in the viewer section. As you can see, this patient presents with ground glass opacities bilaterally and consolidative areas alike, which is a typical case of COVID-19.

Then go to the Segment section, and add three segments, one for the total lung volume, one for the ground glass opacities, and one for the consolidative areas. Go to the Threshold instrument, and set the upper threshold to minus 250 Hounsfield unit. Sort to select only air and leaving out soft tissues, bones, and consolidative areas alike, which the system cannot discern automatically from the soft tissues, then click to apply.

Then, you want to remove the air outside the patient, and to do so, use the Island tool and go to Keep Only Selected Island. Then click inside the patient. You see the air outside have been removed.

Now, you want to add the consolidative areas to the total lung volume. To do so, use the Paint instrument with the Sphere brush setting, adjusting the size to your needings and being careful to avoid soft tissue. With the Sphere brush, you paint up and down with respect to the slice you're working on, which speeds up the process.

Now, you want to segment the consolidative areas. To do so, be sure to be working inside the total lung volume without a variety of other segments. Using the Threshold instrument, raise the upper threshold to be sure to include soft tissues and consolidations, and set the lower threshold to minus 250.

Notice that you're selecting both soft tissues, vessels, and consolidation, but when you click Apply, only consolidative areas have been selected. Now, to segment the ground glass areas, you select the ground glass segment. Keeping the settings from the previous segmentation, set the threshold from minus 750 up to minus 250 Hounsfield unit.

Notice that these value may vary according to the CT scan settings. Always adjust according to the Fleischner Society definition of ground glass. Choose an opacity with that visible vessel in sight.

Then, click Apply. You'll notice that also interstitial spaces have been selected. To adjust, go to the Smooth instrument.

Select whichever seat fits better the construction algorithm of your CT scan. And as you can see, the interstitial spaces have been removed. You can check the results of your segmentation going to Segment Editor, and you can view a 3D model of the segments you have defined.

You can use the slider on the left side to adjust the transparency of the model. 3D View can be used to confirm the correct segmentation process. We now have three distinct segments, one for the total lung volume, and the other two for the typical findings of COVID-19 consolidations and ground glass areas.

You can extract some statistics from the segment you have defined due to the SegmentStatistics module. Remove the Scalar Volume option, and then you have a table expressing the volume of the total lung and the consolidative areas and the ground glass areas. You can use this to calculate the severity of the disease.

You can also export the segment you have defined. You can go to the Data section. Then, make visible only the segment you want to export, clicking on the eye icon.

Then right-click and export binary labelmap only visible. You can remain the labelmap as you prefer. You have to repeat the process for every segment you have defined.

By doing so, you now have three labelmaps corresponding to the segment you have defined. This can be exported in whichever file type you prefer or suits best for further analysis.Furthermore. With these methods, researchers can rapidly obtain an accurate segmentation for pathological findings of COVID-19 for which specific features can be extracted.

Providing such knowledge to the scientific community, accurate, fully automated segmentation methods can be developed and precise prognostic factors be determined.