The age of these 38 patients was 2.00 ± 0.52 years. The youngest patient was 5 months old, and the oldest was 10 years and 10 months old. There were 31 children below 3 years of age. CMT was on the left side in 14 patients and on the right side in 24 patients. The mean cross-sectional area of the SCM was significantly larger on the affected side than on the normal side (1.86 ± 0.69 cm2 vs. 1.17 ± 0.27cm2, P < 0.001; Table 1).

The affected SCM showed abnormal echoes with different degrees of hyperechoic areas (Fig. 2). A local mass-type lesion was seen in the SCM in 8 patients and uniform thickening of SCM in 30 patients. Among the 30 patients with uniform thickening of SCM, the maximum thickness of the muscle was significantly larger on the affected side than on the normal side (7.94 ± 2.36 mm vs. 4.85 ± 1.09 mm, P < 0.001). Among the 8 patients with local mass-type lesion, the mean volume of the mass was 7.184 ± 4.756 cm3 (Supplementary Table 1).

A patient with right congenital muscular torticollis. The patient has limited head rotation to the right but not to the left (a, b). There is limited head flexion to the left side but not to the right side (c, d). The patient’s head is tilted to the right under normal conditions (e). Ultrasound shows significantly thicker sternocleidomastoid muscle on the affected side than on the healthy side (f)

In T1WI of the 38 patients, the mean signal value in the ROI of the affected side is 196.40 ± 47.14, the mean signal value in the ROI of the healthy side is 258.47 ± 44.66, the mean signal value in the ROI was significantly lower on the affected side than on the healthy side (P < 0.001). In T2WI of the 38 patients, the mean signal value in the ROI of the affected side is 145.97 ± 51.61, the mean signal value in the ROI of the healthy side is 211.10 ± 43.65,the mean signal value in the ROI was significantly lower on the affected side than on the healthy side (P < 0.001).In T1 mapping of the 20 patients, the mean signal value in the ROI of the affected side is 1422.15 ± 136.79, the mean signal value in the ROI of the healthy side is 1671.05 ± 160.29,the mean signal value in the ROI was significantly lower on the affected side than on the healthy side (P < 0.001).

In addition, 20 children (53%) showed one or more T1WI and T2WI hypointensity areas in the ROIs of the affected SCMs (Fig. 3). The signal value in T1-mapping images was negatively correlated with the degree of fibrosis. The mean value of T1-mapping was significantly lower on the affected side than on the healthy side (P < 0.001; Supplementary Table 2). The acquired muscle texture features are divided into four categories: Histogram, GLCM, RLM, and Wavelet transform. (Supplementary Table 5).

MRI imaging data and muscle texture data. The region of interest in the sternocleidomastoid of both sides was manually outlined, and T1WI, T2WI, T1-mapping, and Q-dixon sequences were obtained (a–d, respectively). The lowest signal area is seen in the affected SCM (e–h)

Masson staining showed that the percentage of fibrosis is 0.67 ± 0.05. HE staining showed fatty infiltration grade ranging from 0 to 4. While 10 children (26%) had grade 0 fatty infiltration, 28 (74%) had visible fatty infiltration (Fig. 4; Supplementary Table 3).

Histopathology results. Comparison of pathological examination results between platysma and affected sternocleidomastoid muscle. The platysma muscle is seen to be red during the operation (a). Hematoxylin and eosin (HE) staining shows only occasional adipocytes (b, c), and Masson staining shows only mild fibrosis (d, e). In contrast, the affected SCM has a white color (f). HE staining shows abnormal infiltration by adipocytes (green arrow; g, h). Masson staining shows that the fibrosis ratio is higher in the sternocleidomastoid muscle than in the platysma (i, j)

After dimensionality reduction and modeling of statistical data, three indicators—S(2,2)SumAverg, S(3,3)SumVarnc, and T2WI extreme difference—were selected for incorporation into the multivariate conditional logistic regression model. The constructed model had significant diagnostic value for CMT (P < 0.05). High S(2,2)SumAverg was a significant risk factor for presence of CMT (OR = 1.27, 95% CI 1.07–1.50), while high S(3,3)SumVarnc indicated low probability of the disease (OR = 0.91, 95% CI 0.84–0.99; Table 2).

An unconditional multivariate logistic regression diagnostic model incorporating the above three indicators was constructed. This model was similar to those of the conditional multivariate logistic regression model (Table 3). The Hosmer–Lemeshow test showed that the model had a good fit (χ2 = 9.2752, P = 0.3196), indicating that the model was close to the actual identification model. The calibration curve for the Hosmer–Lemeshow test (Fig. 5A) showed a mean square error of 0.02.

Results of Calibration curve of the model and ROC curve of the model. Calibration curve of the model (a). ROC curve of the model (b)

Receiver operating curve (ROC) analysis of the model (Fig. 5B) identified the optimal cutoff score of the model to be 0.366; the area under the curve (AUC) of this model was 0.8283 (95% confidence interval 0.7345–0.922), and the sensitivity and specificity were 0.6842 and 0.8684, respectively.

Finally, a predictive nomogram was constructed for clinical use (Fig. 6).

The fatty infiltration grade on HE-stained specimens was positively correlated to the mean value, standard deviation, and maximum value of the Q-dixon sequence on the affected side (correlation coefficients were 0.65, 0.59, and 0.58, respectively, P < 0.05). The S(2,2)SumAverg and Q-dixon-Dixon sequence [(mean value of affected side − mean value of healthy side)/mean value of healthy side] were negatively correlated with fatty infiltration grade (correlation coefficient was − 0.57, P < 0.05). S(2,2)SumAverg was negatively correlated with the standard deviation of the T1-mapping sequence on the affected side (correlation coefficient was − 0.50, P < 0.05). S(3,3)SumVarnc was positively correlated to the minimum value of the T1-mapping sequence on the affected side (correlation coefficient was 0.47, P < 0.05) (Supplementary Table 4).