Comparison of PET and CT radiomics for prediction of local tumor control in head and neck squamous cell carcinoma

Figures & data

Table 1. Detailed patients characteristic.

	Training cohort	Validation cohort
Number of patients	121	51
Number of recurrences	35 (29%)	15 (29%)
Median follow-up [months]	64	22
Age [years]^a	59 (34–73)	58 (47–75)
HPV status
p16 positive	25 (21%)	28 (55%)
p16 negative	37 (31%)	23 (45%)
not known	59 (48%)	0 (0%)
Tumor stage
T1/T2	45 (37%)	6 (12%)
T3/T4	76 (63%)	45 (88%)
Nodal stage
N0	26 (22%)	9 (18%)
N1	10 (8%)	3 (6%)
N2	81 (67%)	37 (72%)
N3	4 (3%)	1 (2%)
Tumor site
Oropharynx	80 (66%)	28 (55%)
Hypopharynx	24 (20%)	8 (16%)
Larynx	10 (8%)	7 (13%)
Oral cavity	7 (6%)	8 (16%)

^aMinimum and maximum age.

Table 2. The scanning parameters for training and validation cohorts and two image modalities.

Modality	Scanning parameters	Training cohort	Validation cohort
CT	Scanners	Siemens SOMATOM Volume zoom (n = 83) Siemens SOMATOM Definition AS (n = 23) GE Discovery STE (n = 15)	Siemens SOMATOM Definition AS (n = 51)
	Slice thickness [mm]	1.25–3.30	2
	In-plane resolution [mm]	0.85–1.95	0.98–1.56
	kV	120;140	120
	mAs	60–450	183–450
PET	Scanners	GE Discovery STE (n = 51) GE Discovery 690 (n = 3) GE Discovery HR (n = 24) GE Discovery LS (n = 20) GE Discovery RX (n = 23)	GE Discovery STE (n = 35) GE Discovery 690 (n = 7) GE Discovery RX (n = 9)
	Reconstruction	2D (n = 61) 3D (n = 60)	2D (n = 0) 3D (n = 51)
	Slice thickness [mm]	3.3; 4.3	3.3
	In-plane resolution [mm]	2.7–5.5	2.7–5.5
	Administered 18F-FDG activity [MBq]	351 (237–513)^a	326 (178–406)^a

^aMinimum and maximum administered activity.

Figure 1. Radiomics-based local tumor control prognostic models: (a) CT-based radiomics, (b) PET-based radiomics, (c) CT- and PET-based radiomics. Tumor control curves split significantly (G-rho test p-value < 0.05) in both training and validation cohorts based on the optimal sensitivity-specificity thresholds at 18 months.

Figure 2. The comparison of risk group stratification in the validation cohort by CT- and PET-based radiomics models (a) the repeatability of patients’ ranking based on CT and PET radiomics models and differences in risk group stratification, (b) the relation between observed and radiomics model-based estimated probabilities of local tumor control at 18 months for risk groups defined using the radiomics models.