Development and validation of a combined model based on dual-sequence MRI radiomics for predicting the efficacy of high-intensity focused ultrasound ablation for hysteromyoma

Figures & data

Figure 1. Flowchart of patient recruitment with inclusion and exclusion criteria (a, b).

Figure 2. Flowchart showing the process for the development of radiomics.

Table 1. Clinical–radiological features of patients in the training, validation and independent test datasets.

Clinical–radiological features	Training and validation cohorts (n = 172)			Test cohort (n = 40)			p
	Insufficient ablation (n = 77)	Sufficient ablation (n = 95)	p	Insufficient ablation (n = 13)	Sufficient ablation (n = 27)	p
Age (years)^c	45 (42–49)	45 (41–47)	0.292a	43 (36–47)	36 (33–43)	0.189a	<0.001^a
Volume^c	37983.78 (23710.59–90801.70)	73943.71 (45071.00–130346.97)	<0.001^a	47890.66 (23388.17–108060.77)	53925.14 (25359.10–107848.31)	0.864^a	0.683^a
Type			0.770b			0.574^b	0.990^b
Subserous	13 (16.88)	8 (8.42)		0 (0.00)	2 (7.41)
Intramural	63 (81.82)	85 (89.47)		11 (84.62)	22 (81.48)
Submucous	1 (1.30)	2 (2.11)		2 (15.38)	3 (11.11)
T2WI signal intensity			<0.001^b			0.950^b	<0.001^b
Hyperintensity	34 (44.16)	18 (18.95)		6 (46.15)	3 (11.11)
Isointensity	13 (16.88)	10 (10.52)		4 (30.77)	12 (44.44)
Hypointensity	30 (38.96)	67 (70.53)		3 (23.08)	12 (44.44)
Enhancement degree			0.004^b			0.989^b	0.752^b
Mild	25 (32.47)	48 (50.53)		3 (23.08)	15 (55.56)
Moderate	27 (35.06)	35 (36.84)		8 (61.54)	4 (14.81)
Obvious	25 (32.47)	12 (12.63)		2 (15.38)	8 (29.63)
Enhancement signal homogeneity			0.766^b			1.000^b	0.622^b
Homogeneous	21 (27.27)	24 (25.26)		4 (30.77)	8 (29.63)
Inhomogeneous	56 (72.73)	71 (74.74)		9 (69.23)	19 (70.37)
CDFI signal			<0.001^b			<0.001^b	0.054^b
Small amount	60 (77.92)	93 (97.89)		5 (38.46)	26 (96.30)
Slightly abundant	17 (22.08)	2 (2.11)		8 (61.54)	1 (3.70)

Data are shown as number of patients and percentage in parentheses, unless otherwise stated.

^a p Values obtained using Wilcoxon’s rank-sum test.

^b p Values obtained using chi-squared test or Fisher’s exact test.

^c Data are medians and quartiles in parentheses.

Table 2. Univariate and multivariate analyses results of clinical–radiological features.

Clinical–radiological features	Univariate		Multivariate
	OR (95% CI)	p Value	OR (95% CI)	p Value
Age	0.453 (0.080–2.279)	0.348
Volume	3.742 (0.484–80.002)	0.283
Type	1.724 (0.923–3.348)	0.093
T2WI signal intensity	1.106 (1.047–1.171)	<0.001	1.064 (1.002–1.131)	0.044
Enhancement degree	0.810 (0.686–0.952)	0.012	0.851 (0.712–1.013)	0.072
Enhancement signal homogeneity	1.072 (0.676–1.697)	0.766
CDFI signal	0.024 (0.002–0.155)	0.001	0.045 (0.003–0.325)	0.007

Figure 3. ROC curves of clinical–radiological models, radiomics models and combined models in the training (a, c, e) and validation (b, d, f) cohorts.

Table 3. Performance of clinical–radiological, radiomics and combined models in the training and validation cohorts.

Model	AUC (95% CI)	Sensitivity	Specificity	Accuracy	Precision	F1 score
Clinical–radiological model	0.693 (0.611–0.775)|0.689 (0.525–0.853)	0.800|0.800	0.481|0.482	0.657|0.657	0.655|0.655	0.720|0.720
Radiomics model	0.803 (0.726–0.881)|0.794 (0.631–0.954)	0.782|0.758	0.705|0.676	0.747|0.721	0.765|0.744	0.773|0.751
Combined model	0.841 (0.772–0.909)|0.834 (0.691–0.971)	0.837|0.832	0.737|0.716	0.792|0.780	0.797|0.785	0.817|0.808

Training cohort|validation cohort.

Table 4. Comparison of models in the training and validation cohorts.

Model	Group	NRI (95% CI)	p (NRI)	IDI (95% CI)	p (IDI)
Combined model vs. clinical–radiological model	Training cohort	0.314 (0.144–0.484)	<0.001	0.104 (0.071–0.137)	<0.001
	Validation cohort	0.253 (–0.093 to 0.598)	0.152	0.141 (0.075–0.208)	<0.001
Combined model vs. radiomics model	Training cohort	0.098 (–0.016 to 0.212)	0.092	0.029 (0.012–0.045)	<0.001
	Validation cohort	0.119 (–0.098 to 0.337)	0.282	0.053 (0.017–0.089)	0.004

Net reclassification improvement (NRI) >0 and IDI >0 were positive improvement, indicating that the predictive ability of the new model was better than the old one.

Figure 4. The combined nomogram for predicting the efficacy of HIFU for hysteromyoma (a). The developed nomogram based on the combined model to predict the risk of insufficient HIFU ablation. T2WI signal intensity: 1, hyperintensity; 2, isointensity; 3, hypointensity; CDFI signal: 1, small amount; 2, slightly abundant. Calibration curves for the nomogram in the training and validation cohorts (b, c). The black dashed line represents the ideal prediction, and the green, red and purple lines represent the predictive ability of clinical–radiological models, radiomics models and combined models. The closer they fit to the dashed green line, the greater the prediction accuracy of the model. Decision curve analysis for the nomogram (d). The black line represents the net benefit of assuming no hysteromyoma patients receive HIFU ablation. The grey line is the net benefit of assuming that all hysteromyoma patients receive HIFU ablation. The green line, red line and purple line represent the expected net benefit of predicting stroke outcome using the clinical–radiological models, radiomics models and combined models, respectively.

Data availability statement

The data used to support the findings of this study are available from the corresponding author upon request