A new automated delineation method for SPECT lung scans using adaptive dual‐exponential thresholding

Abstract

An accurate method for delineating lung contours in single photon emission computed tomography (SPECT) is critical to respiratory studies such as pulmonary embolism (PE) and respiratory aerosol deposition. Current delineation methods are not adaptive in nature and may require a priori information on lung volumes. We have developed a dual-exponential thresholding method that solely requires SPECT scans, and is fast, accurate, and adaptive in nature. A dataset consisted of 90 patients was studied retrospectively. While most were suspected of PE, other pulmonary disorders were also present. SPECT ventilation scans were obtained after inhalation of *40 MBq of 99mTc-Technegas. Examination of the corresponding natural logarithmically transformed histograms revealed dual exponential functions like pattern. Adaptive thresholds were found as the intercept between the two exponential functions. 350 Monte Carlo simulations representative of normally-ventilated scans with varying counts were used to quantitatively measure and evaluate the method. Our method correctly delineated lung volumes to above 97% agreement for 310 of the 350 Monte Carlo simulations with maximum count values greater than or equal to 31. The 40 low-count simulations were included purely as extreme cases. For real patient scenario, an experienced physician was called upon to visually delineate a sample of 50 actual patient scans, and these delineated lung volumes were used as gold standard and compared with those delineated using our method. Our results had an average of 96% agreement.

Single-Photon Emission Computed Tomography (SPECT)

“…Congruency is calculated on four datasets previously used to evaluate our dual exponential thresholding method [9] and subsequent coupling with planar active contours [10] respectively: (1) two sets of 50 randomly selected simulations and real subjects with normal maximum and/or total count values, and (2) 90 simulations with low maximum and/or total count values and 35 real subjects with similarly ranged maximum/total count values. as the known phantom volume of 61,660 voxels.…”

“…This approach however is subject to the presence of localised high deposition of radioactive agents known as "hotspots". While we have overcome this limitation with our new method of dynamic dual exponential thresholding [9] and the subsequent coupling with traditional planar active contours [10], we report our findings on the implementation of true three-dimensional (3D) active contours for SPECT lung delineation in this paper. Since the active contour method was first proposed in 1988 [11], and although a lot of research has been carried out into improving the method in terms of speed [12][13][14][15], stability [16][17][18], and convergence [19][20][21][22][23], most of these methods were applied to planar images.…”

Automatic delineation and quantification of pulmonary vascular obstruction index in patients with pulmonary embolism using Perfusion SPECT-CT: a simulation study

“…The best threshold was much higher as compared with our study (40% vs 15%) for Derlin, but closer for Seiffert [ 24 ] (18% for left lung and 21% for right lung), illustrating the lake of robustness of the method, especially if there are hot spots. Wang et al [ 25 – 27 ] proposed three segmentation methods (one threshold based method and two adaptive contouring methods), tested on Monte-Carlo simulations of homogeneous distributions of radioactivity within the lungs. The results, expressed as the ratio of the intersection and the union of the ground truth and the segmented volume, were variable depending on the count rates.…”

Delineating low‐count defective‐contour SPECT lung scans for PE diagnosis using adaptive dual exponential thresholding and active contours

“…Such an approach is prone to either under-delineation due to low maximum count values used in the 25% count or removal of true lung tissues due to ''hotspots''-localised high deposition of radioactive agents. In our previous study, although we supported basic delineation using 10 and 25% maximum count values on real subject scans and phantom studies generated using Monte Carlo simulation respectively, we provided a more rigorous and accurate adaptive thresholding method, which we called dual exponential thresholding (Wang and Yan, 2007). The method first applied natural logarithmic transformation to the SPECT ventilation scan histogram, displaying three exponential-like components: a fast section representing background noise, a relatively horizontal section representing most of the lung tissues, and a final less distinctive section representing high values and noisy hotspots.…”