Fully automatic lung lobe segmentation using V/Q SPECT/CT images

Ziel/Aim Ventilation/Perfusion (V/Q) SPECT largely replaced planar V/Q scintigraphy in daily practice for the diagnosis of pulmonary embolism (PE). The identification of the lung lobes has important applications in disease assessment. However the lobe separation is usually done with manual contouring or semi-automatic tools. The fuzzy appearance of fissures on low-dose CT images makes the computer-aided lobe detection a challenging task. On the usually used low-dose CT fissures frequently appear partially and blurred due to partial volume effects and patient motion. This work presents a fully automatic approach for lung lobe segmentation focusing on the V/Q SPECT clinical protocols using low-dose CT images.

Methodik/Methods Our proposed method is a combination of three main approaches: airways-based labeling and partitioning, vessel-based approach and detection of the visible fissures borders. We have put emphasis on the separation of the lobes even if no or very weak fissure borders are presented in the low-dose CT slices. Our lung lobe segmentation method needs a previously segmented lung mask.

Ergebnisse/Results The method was tested on 10 V/Q SPECT/CT images (120 KeV, 50-100 mA, slice thickness: 2.5 mm) from the daily clinical routine acquired on the Mediso AnyScan SPECT/CT system. The fully automatic lung lobe segmentation succeeded in every case (10/10). The results were validated by two specialists in nuclear medicine. The generated lung lobes can be fused with the SPECT images to determine the V/Q mismatches at lobar level.

Schlussfolgerungen/Conclusions The fully automatic lung lobe segmentation is feasible using low-dose CT images, even if there are no or weak fissure boundaries.