Speakers
Description
I. BACKGROUND
Pulmonary Embolism (PE) is reported to be one of the most
common cardiovascular diseases. It is caused by a blood clot
that develops in a blood vessel elsewhere in the body and
travels to an artery in the lung, forming a blockage [1].
Lung V/Q (ventilation/perfusion) SPECT is one of an es-
tablished diagnostic imaging test for suspected PE. The idea
behind this test is to administer patient the radioactive tracer
intravenous or by inhalation and use a gamma camera to detect
the radiation emitted. The regular CT is also often performed
to improve diagnosis sensitivity and specificity [2].
II. PURPOSE
The goal is to develop an algorithm estimating lung volumes
from Perfusion SPECT and CT examinations and calculate the
ratio VP /VCT (Perfusion lung volume to CT lung volume).
The assumption is that a low ratio may indicate the presence
of regions with no perfusion, suggesting the patient may be
suffering from PE. This algorithm may improve the sensitivity
and specificity of PE diagnosis and introduces an automatically
calculated metric quantifying the severity of potential PE.
III. MATERIALS AND METHODS
During our studies, we developed a method for calculating
lung volumes using two complementary imaging modalities:
SPECT and CT. Our dataset comprised 5 patients — 3
diagnosed with Pulmonary Embolism (PE) and 2 healthy
controls. Both scans were available for each patient, acquired
through our collaboration with the Warsaw Military Institute
of Medicine.
Data processing involved analysis of DICOM files contain-
ing both CT and SPECT images along with their associated
metadata. For segmentation, we employed distinct approaches
tailored to each modality: SPECT images were segmented
using a standardized fixed threshold method, while CT im-
ages were processed using an established machine learning
algorithm based on transfer learning of U-net architecture,
implemented through the lungmask [3] library. We improved
the CT segmentation process by converting pixel values to
Hounsfield Units (HU) as a crucial preprocessing step. Lung
volumes for both imaging modalities were calculated using
the physical voxel dimensions from the image metadata. This
dual-modality approach ultimately enabled us to calculate the
Perfusion/CT ratio based on the segmented volumes.
IV. RESULTS
The developed algorithm successfully distinguished be-
tween PE-diseased and healthy patients. The maximum vol-
ume ratio observed in the diseased group was 0.695, while
the minimum value in the healthy group was 0.869.
V. CONCLUSION
Although the algorithm effectively differentiated between
the two groups, the limited sample size suggests the need for
further research and additional data to validate the findings,
determine a reliable threshold, and establish fundamental
metrics such as sensitivity and specificity.
REFERENCES
[1] A. C. Clark, J. Xue, and A. Sharma, ‘Pulmonary Embolism: Epi-
demiology, Patient Presentation, Diagnosis, and Treatment’, Journal
of Radiology Nursing, vol. 38, no. 2, pp. 112–118, Jun. 2019, doi:
10.1016/j.jradnu.2019.01.006.
[2] M. Bajc, J. B. Neilly, M. Miniati, C. Schuemichen, M. Meignan, and
B. Jonson, ‘EANM guidelines for ventilation/perfusion scintigraphy:
Part 1. Pulmonary imaging with ventilation/perfusion single photon
emission tomography’, Eur J Nucl Med Mol Imaging, vol. 36, no. 8,
pp. 1356–1370, Aug. 2009, doi: 10.1007/s00259-009-1170-5.
[3] J. Hofmanninger, F. Prayer, J. Pan, S. R¨ohrich, H. Prosch, and G. Langs,
‘Automatic lung segmentation in routine imaging is primarily a data
diversity problem, not a methodology problem’, European Radiology
Experimental, vol. 4, no. 1, p. 50, Aug. 2020, doi: 10.1186/s41747-020-
00173-2.
