Clinical validation of an AI-based automatic quantification tool for lung lobes in SPECT/CT

EJNMMI Physics

Table 1 Average [range] relative distribution (%) of the ventilation radiotracer in the right and left lung lobes and total lungs, as determined by using the three quantification methods

Quantification method	Right lung lobe values			Left lung lobe values		Total lung values
Quantification method	Superior lobe	Middle lobe	Inferior lobe	Superior lobe	Inferior lobe	Right lung	Left lung
Planar scintigraphy	10.7 [2.4–21.6]^a	31.2 [8.5–40.5]^a	13.1 [1.2–34.1]^a	24.3 [10.9–39.2]	20.7 [10.2–47.6]	55.0 [13.3–78.8]	45.0 [21.2–86.7]
SPECT AutoLung3D	20.9 [6.0–38.0]	9.5 [1.0–19.0]	23.8 [6.0–43.0]	25.5 [15.0–53.0]	20.3 [3.0–34.0]	54.2 [13.0–79.0]	45.8 [21.0–87.0]
SPECT manual segmentation	18.8 [4.7–39.3]	9.7 [1.6–21.2]	25.4 [9.3–45.1]	23.4 [0^b–48.9]	22.7 [2.1–44.3^b]	53.9 [21.7–78.3]	46.1 [21.7–78.3]
p*	< 0.001	< 0.001	< 0.001	0.433	0.343	0.824	0.828

Bold p-values mean a statistically significant difference between the quantification values obtained with the manual and automated methods
The data are shown as mean [range]
^aThe automated and manual values both differ significantly from the planar scintigraphy value
^bOne patient had had a lobectomy, which was taken in account only on manual segmentation as it was not visible on planar images and the lobectomized lung was not accurately segmented by the AutoLung3D algorithm
*Determined by ANOVA