Development of a Clinical-Radiomics Nomogram That Used Contrast-Enhanced Ultrasound Images to Anticipate the Occurrence of Preoperative Cervical Lymph Node Metastasis in Papillary Thyroid Carcinoma Patients

Figures & data

Table 1 Patients’ Characteristics of Training Set and Validation Set

Characteristic	Training Set (n = 282)		Validation Set (n = 141)		P-value
	n	%	n	%
Age					0.212
<55 years	185	65.60%	101	71.63%
≥55 years	97	34.40%	40	28.37%
Gender					0.23
Female	179	63.48%	81	57.45%
Male	103	36.52%	60	42.55%
Primary site					0.103
Left lobe	137	48.58%	84	59.57%
Right lobe	115	40.78%	45	31.91%
Isthmus	30	10.64%	12	8.51%
Tumor location					0.06
Intra-thyroidal	43	15.25%	32	22.70%
Sub-capsular	239	84.75%	109	77.30%
Tumor size					0.63
≤10 mm	139	49.29%	66	46.81%
>10 mm	143	50.71%	75	53.19%
Lymph node metastasis					0.891
Yes	138	48.94%	68	48.23%
No	144	51.06%	73	51.77%
Echogenicity					0.475
Iso/hyperechoic	35	12.41%	21	14.89%
Hypoechoic	156	55.32%	65	46.10%
Marked hypoechoic	91	32.27%	55	39.01%
Aspect ratio > 1					0.945
Absent	125	44.33%	63	44.68%
Present	157	55.67%	78	55.32%
Margin					0.075
Smooth	21	7.45%	11	7.80%
Ill-defined	39	13.83%	9	6.38%
Irregular	222	78.72%	121	85.82%
Microcalcification					0.509
Absent	95	33.69%	43	30.50%
Present	187	66.31%	98	69.50%
Enhancement pattern					0.64
Hyper-enhancement	21	7.45%	8	5.67%
Iso-enhancement	59	20.92%	34	24.11%
Hypo-enhancement	202	71.63%	99	70.21%
US-reported LN status					0.17
Positive	43	15.25%	29	20.57%
Negative	239	84.75%	112	79.43%
BMUS Radscore, Median (Interquartile range)	−0.38 (−0.70, −0.06)		−0.32 (−0.79, 0.15)		0.731
CEUS Radscore, Median (Interquartile range)	−0.51 (−1.11, 0.09)		−0.39 (−0.83, 0.05)		0.292

Figure 1 The flowchart of the training and validating process for clinical-radiomics nomogram modeling.

Abbreviations: BMUS, B-mode ultrasound; CEUS, contrast-enhanced ultrasound; ICC, interclass correlation coefficient; mRMR, minimum redundancy maximum relevance; LASSO, least absolute shrinkage and selection operator; AIC, Akaike information criterion; Radscore, radiomics score.

Figure 2 The method of choosing radiomics features from the training set using the LASSO algorithm. (A and C) show the LASSO coefficient profiles for the BMUS and CEUS features, respectively. In (B and D), it is shown how to apply minimal criteria analysis and 10-fold cross-validation to get the ideal penalization coefficient lambda () for the BMUS and CEUS LASSO models, respectively. The lambda values chosen based on the minimal criterion and one standard error of the minimum criteria are shown as dotted vertical lines.

Table 2 The Results of Stepwise Multivariate Analyses for Prediction of LNM

Characteristics	Odds Ratio (95% CI)	P-value
Clinical model
Gender	2.33 (0.78, 6.05)	0.067
Age	0.27 (0.05, 0.86)	0.038
Tumor size	2.03 (0.95, 4.72)	0.051
US-reported LN status	4.17 (1.10, 12.86)	0.006
Clinical-radiomics model
Gender	1.22 (0.86, 3.29)	0.300
Age	1.39 (1.01, 3.25)	0.015
CEUS Radscore	6.12 (1.79, 11.45)	0.034
US-reported LN status	2.63 (1.80, 3.78)	0.005

Figure 3 The ROC curves of the clinical model and clinical-radiomics model in the training set (A) and validation set (B).

Abbreviation: ROC, receiver operating characteristic.

Figure 4 A radiomic nomogram for evaluating cervical LNM based on the clinical-radiomics model (A). Clinical-radiomics nomogram calibration curves in the training set (B) and validation set (C).

Figure 5 Decision curve of clinical model and clinical-radiomics model. The decision curve analysis (DCA) measures the net benefit (y-axis) versus the model’s high-risk threshold (x-axis) for different models.