Clinical Validation and Extension of an Automated, Deep Learning–Based Algorithm for Quantitative Sinus CT Analysis

Article Figures & Data

Download figure
Open in new tab
Download powerpoint
FIG 1.
3D reconstructed sinus CT scan with bony overlay (gray) and segmentation of the paranasal sinuses (colored) produced by the algorithm.
Download figure
Open in new tab
Download powerpoint
FIG 2.
Illustration demonstrating application of osteitis segmentation rind. Left, Coronal CT with total GOSS score of 32. Right, Segmentation rind overlaid in red. Forty-eight percent of the rind is occupied by CT voxels corresponding to bone.
Download figure
Open in new tab
Download powerpoint
FIG 3.
Flow diagram demonstrating the cohort-selection process.
Download figure
Open in new tab
Download powerpoint
FIG 4.
Scatterplot demonstrating algorithm-generated %SO versus LMS. The gray shaded area represents a 95% confidence interval for the regression line (ρ = 0.85, P < .001).
Download figure
Open in new tab
Download powerpoint
FIG 5.
Algorithm-derived %OST versus total GOSS as stratified by surgical status. Overall ρ = 0.48 (P < .05). For surgically naïve patients, ρ = 0.29 (P = .08), and for postoperative subjects, ρ = 0.59 (P < .001).

Table 1:

Demographics of the study population

Cohort Features	n = 88 (%, or range)
Female (%)	52 (59)
Avg. age (range) (yr)	48.5 (22–78)
White (%)	70 (80)
Polyps (%)	59 (67)
Asthma (%)	53 (60)
AERD (%)	8 (9)
Never smoker (%)	55 (62)
Prior surgery (%)	48 (55)
Avg. absolute serum eosinophils × 10⁹/L	0.31 (0–2.6)
Avg. tissue eosinophils per HPF	80 (0–413)
Avg. SNOT-22	49.6 (15–95)
Avg. LKS	6.8 (1–12)
Avg. LMS	13.5 (2–24)
Avg. GOSS	2.7 (0–35)
Avg. sinus cavity volume (mL)	60.8 (24.8–109.0)
Avg. %SO	60.1 (20.7–99.6)
Avg. mHU	10.5 (−139.2 to +57.4)
Avg. %OST	41.4 (32.1–52.2)

Note:—AERD indicates aspirin-exacerbated respiratory disease; HPF, high-power field; Avg., average.

Table 2:

CT scanner metrics and acquisition protocols used for the included scans

Model (No.)	Reconstruction Kernel (No.)	Tube Potential (No.)	Tube Current (No.)	Pitch (No.)	Axial Section Thickness (No.)	Axial Section Spacing (No.)
Siemens (n = 77)
Sensation 64 (42)	H31s (2)	100 kVp (2)	Modulated (22)	0.7 (43)	1.0 mm (77)	0.6 mm (2)
Definition (4)	H60f (2)	120 kVp (75)	91 mA (41)	0.8 (19)		0.9 mm (63)
Definition AS (3)	H60s (1)		100 mA (8)	0.9 (14)		1.0 mm (12)
Definition AS+ (4)	H70h (54)		108 mA (6)	1.0 (1)
Definition Flash (18)	J40s\\2 (1)
Biograph 40 (6)	J70h\\2 (17)
GE Healthcare (n = 5)
LightSpeed Pro 16 (1)	Bone (5)	120 kVp (5)	110 mA (1)	0.50–0.75 (2)	0.625 mm (3)	0.625 mm (3)
LightSpeed VCT (1)			150 mA (1)	0.90–1.0 (3)	1.25 mm (2)	1.0 mm (1)
Optima CT540 (2)			160 mA (1)			1.25 mm (1)
Optima CT660 (1)			>200 mA (2)
Philips Healthcare (n = 4)
Brilliance 64 (4)	YC (3)	120 kVp (4)	85 mA (1)	1.0 (4)	1.0 mm (4)	0.8 mm (1)
	YD (1)		119 mA (2)			0.9 mm (3)
			170 mA (1)
Toshiba/Canon (n = 2)
Aquilion ONE (1)	FC30 (2)	120 kVp (2)	150 mA (1)	1.0 (2)	1.0 mm (2)	0.8 mm (1)
Aquilion Prime (1)			200 mA (1)			1.0 mm (1)

Table 4:

Correlation (ρ) values for percentage sinus opacification stratified by sinus cavity and clinical parameters

Clinical Parameter	Left Maxillary	Right Maxillary	Anterior Ethmoid	Posterior Ethmoid	Frontal	Sphenoid
LKS	0.43^{^a}	0.34^{^b}	0.64^{^a}	0.56^{^a}	0.53^{^a}	0.35^{^b}
SNOT-22	0.10	0.06	0.06	−0.03	0.10	−0.10
Absolute serum eosinophils	0.03	0.15	0.23^{^b}	0.26^{^b}	0.21	0.18
Tissue eosinophil count	0.13	0.21	0.37^{^a}	0.39^{^a}	0.23^{^b}	0.11