Patient’s characteristics

As shown in Table 1, the patients with AT had older age (AT: 47 ± 19 years; non-AT: 40 ± 15 years; p = 0.014), shorter pain duration (AT: 6 ± 6 days; non-AT: 10 ± 9 days; p < 0.001), and higher neutrophil-lymphocyte ratio (AT: 6.9 ± 6; non-AT: 4.4 ± 3.8; p < 0.001). The four most frequently seen lesion types in the AT group were ovarian cyst (32%, 23/73), serous cystadenoma (18%, 13/73), teratoma (15%, 11/73), and mucinous cystadenoma (12%, 9/73). The proportions of endometrioma and tubo-ovarian abscess were higher in the non-AT group (endometrioma: 15%, 14/92; tubo-ovarian abscess 20%, 18/92) than those in the AT group (endometrioma 3%, 2/73, p = 0.007; tubo-ovarian abscess: 1%, 1/73, p < 0.001).

Derivation and validation of ΔHUPV-NC

The attenuation of the lesion was measured based on ROI plotting in contrast-enhanced CT images in both non-contrast and portal-venous phases (Figs. 3 and 4).

Fig. 3

Attenuation measurement in contrast-enhanced CT of a patient (age: 53 years) with a chief complaint of abdominal pain for 1 day. Surgically and pathologically confirmed twisted ovarian cyst. a Non-contrast phase (attenuation: 51.5 HU). b Portal-venous phase (attenuation: 63.8 HU). The ΔHUPV-NC is 12.3 HU

Fig. 4

Attenuation measurement in contrast-enhanced CT of a patient (age: 22 years) with a chief complaint of abdominal pain for 4 days. Surgically and pathologically confirmed untwisted ovarian cyst. a Non-contrast phase (attenuation: 25.5 HU). b Portal-venous phase (attenuation: 48.2 HU). The ΔHUPV-NC is 22.7 HU

In the derivation set, the mean attenuation in the non-contrast phase was higher in the AT group (38.8 ± 12.2 HU) than in the non-AT group (29.8 ± 5.5 HU, p < 0.001, Fig. 5a). In contrast, the mean attenuation in the portal-venous phase was lower in the AT group (48.5 ± 16.2 HU) than in the non-AT group (65.6 ± 19.7 HU, p < 0.001, Fig. 5a). The same pattern was identified in the mixed validation set as the mean attenuation in the AT group was higher in the non-contrast phase (AT: 42.5 ± 16.4 HU; non-AT: 30 ± 9.7 HU; p < 0.001; Fig. 5a) but lower in the portal-venous phase (AT: 52.6 ± 18.2 HU; non-AT: 64.1 ± 22.6 HU; p = 0.02, Fig. 5a).

Fig. 5

Attenuation and ΔHUPV-NC in derivation and mixed validation sets. a Attenuation of the twisted vs untwisted lesions in derivation and mixed validation sets. Derivation set: non-contrast phase (torsion: 38.8 ± 12.2 HU; non-torsion: 29.8 ± 5.5 HU, p < 0.001) and portal-venous phase (torsion: 48.5 ± 16.2 HU; non-torsion: 65.6 ± 19.7 HU; p < 0.001). Mixed validation set: non-contrast phase (torsion: 42.5 ± 16.4 HU; non-torsion: 30 ± 9.7 HU; p < 0.001) and portal-venous phase (torsion: 52.6 ± 18.2 HU; non-torsion: 64.1 ± 22.6 HU; p = 0.02). b ΔHUPV-NC (cutoff value: 17.5 HU) of the twisted vs untwisted lesions in derivation and mixed validation sets. Derivation set: (torsion: 9.7 ± 11.4 HU; non-torsion: 36 ± 16.2 HU, p < 0.001). Mixed validation set: (torsion: 10.1 ± 8 HU; non-torsion: 36.2 ± 16.6 HU, p < 0.001). c The receiver operating characteristic curve (ROC) of the ΔHUPV-NC in the derivation (red solid round; torsion: n = 47; non-torsion: n = 58; sensitivity: 96% [95% CI: 86, 99]; specificity: 89% [95% CI: 79, 95]) and mixed validation (blue open square; torsion: n = 26; non-torsion: n = 34; sensitivity: 92% [95% CI: 76, 99]; specificity: 88% [95% CI: 73, 95]) sets

The ΔHUPV-NC was measured by subtracting the non-contrast HU from portal-venous HU. In both derivation and mixed validation sets, the ΔHUPV-NC in the AT group (derivation: 9.7 ± 11.4 HU; mixed validation: 10.1 ± 8 HU) was lower than those in the non-AT group (derivation: 36 ± 16.2 HU, p < 0.001; mixed validation: 36.2 ± 16.6 HU, p < 0.001; Fig. 5b). Once the ROC of ΔHUPV-NC was plotted, the AUC of the derivation set (0.92 [95% CI: 0.86, 0.98]) was comparable to that of the mixed-validation set (0.91 [95% CI: 0.83, 0.99]; p = 0.86; Fig. 5c) [18]. With the cutoff value of 17.5 HU, the sensitivities of ΔHUPV-NC in derivation (96% [95% CI: 86, 99]) and mixed-validation (92% ([95% CI: 76, 99]) sets were comparable (p = 0.61, Fig. 5C and Table S3). The difference in the specificity of ΔHUPV-NC in derivation (89% [95% CI: 79, 95]) and mixed-validation (88% ([95% CI: 73, 95]) sets was insignificant (p = 1; Fig. 5c and Table S4).

In the pooled study sample, the mean attenuation in the non-contrast phase was higher in the AT group (40.1 ± 10.7 HU) than in the non-AT group (29.7 ± 8.3 HU, p < 0.001, Fig. 6a). In the portal-venous phase, the mean attenuation in the AT group (50 ± 16.9 HU) was lower than in the non-AT group (68.4 ± 20.7 HU, p < 0.001, Fig. 6a). The ΔHUPV-NC was lower in the AT group (9.9 ± 10.3 HU) than in the non-AT group (39.3 ± 17.6 HU, p < 0.001; Fig. 6b). The AUC, sensitivity, and specificity of the pooled sample were 0.91 (95% CI: 0.86, 0.96), 95% (95% CI: 87, 98), and 88% (95% CI: 80, 94), respectively (Fig. 6c and Table 3).

Fig. 6

Attenuation and ΔHUPV-NC in the pooled sample. a Attenuation of twisted vs untwisted lesions in the pooled sample. Non-contrast phase (torsion: 40.1 ± 10.7 HU; non-torsion: 29.7 ± 8.3 HU; p < 0.001). Portal-venous phase (torsion: 50 ± 16.9 HU; non-torsion: 68.4 ± 20.7 HU; p < 0.001). b ΔHUPV-NC (cutoff value: 17.5 HU) of twisted vs untwisted lesions in the pooled sample. Torsion: 9.9 ± 10.3 HU; non-torsion: 39.3 ± 17.6 HU (p < 0.001). c The ROC of the ΔHUPV-NC in the pooled sample (torsion: n = 73; non-torsion: n = 92; sensitivity: 95% [95% CI: 87, 98]; specificity: 88% [95% CI: 80, 94])

Table 3 Diagnostic accuracy and logistic regression of CT signs to predict ATDiagnostic performance of ΔHUPV-NC in patients with ultrasonography-unspecified AT

The sensitivity of ΔHUPV-NC ≤ 17.5 HU (95% [95% CI: 87, 98]) was higher than those of the whirlpool sign (37% [95% CI: 27, 49], p < 0.001), peritoneal ascites (42% [95% CI: 32, 54], p < 0.001), lesion wall thickening (44% [95% CI: 33, 55], p < 0.001), intralesional hemorrhage (40% [95% CI: 29, 51], p < 0.001), and fat stranding (44% [95% CI: 33, 55], p < 0.001; Tables 3 and S5–9). The specificity of ΔHUPV-NC ≤ 17.5 HU (88% [95% CI: 80, 94]) was higher than those of peritoneal ascites (71% [95% CI: 61, 79], p = 0.006) and fat stranding (60% [95% CI: 50, 70], p < 0.001), but comparable to those of whirlpool sign (93% [95% CI: 87, 97], p = 0.17), lesion wall thickening (85% [95% CI: 76, 91], p = 0.49), and intralesional hemorrhage (77% [95% CI: 68, 85], p = 0.06 Tables 3 and S10–14).

Eleven and four patients were false positive and false negative for the sign of ΔHUPV-NC ≤ 17.5, respectively (Table S15 and Figs. S1 and S2).

In univariable logistic regression, the AT was associated with the ΔHUPV-NC ≤ 17.5 HU (OR = 127 [95% CI: 39, 417], p < 0.001), whirlpool sign (OR = 8.4 [95% CI: 3.2, 21.8], p < 0.001), lesion wall thickening (OR = 4.4 [95% CI: 2.1, 9.1], p < 0.001), intralesional hemorrhage (OR = 2.2 [95% CI: 1.1, 4.4], p = 0.02), and fat stranding (OR = 1.2 [95% CI: 0.6, 2.2], p = 0.005). In multivariable logistic regression, the AT was only associated with the ΔHUPV-NC ≤ 17.5 HU (OR = 137 [95% CI: 39, 481], p < 0.001; Tables 3 and S16).

Diagnostic performance of ΔHUPV-NC in patients with ultrasound suspicion of AT

In patients with ultrasound suspicion of AT (torsion: n = 37, non-torsion: n = 16), the AUC of ΔHUPV-NC was 0.94 (95% CI: 0.88, 1; Fig. S3). With the cutoff value of 17.5 HU, the sensitivity and specificity were 95% (95% CI: 82, 99) and 81% (95% CI: 54, 96; Fig. S3 and Table S17). The diagnostic accuracy of ΔHUPV-NC ≤ 17.5 HU (88%, [95% CI: 75, 96]) was higher than those of whirlpool sign (68% [95% CI: 54, 80], p = 0.013), peritoneal ascites (53% [95% CI: 39, 67], p < 0.001), lesion wall thickening (66% [95% CI: 52, 78], p < 0.001), intralesional hemorrhage (53% [95% CI: 39, 67], p < 0.001), and fat stranding (57% [95% CI: 42, 70], p < 0.001; Table S17).

Diagnostic performance of ΔHUPV-NC in pooled patients with either ultrasonography-specified or ultrasonography-unspecified AT

Once the patients with either ultrasonography-specified or ultrasonography-unspecified AT were pooled together (torsion: n = 110, non-torsion: n = 108), the AUC of ΔHUPV-NC was 0.92 (95% CI: 0.87, 0.96, p < 0.001; Fig. S4). With the cutoff value of 17.5 HU, the sensitivity and specificity were 95% (95% CI: 89, 98) and 87% (95% CI: 79, 92; Fig. S4), similar to the sensitivity (95% [95% CI: 87, 98], p = 1) and specificity (88% [95% CI: 80, 94], p = 1) in patients with ultrasound-unspecified AT (Tables 3, S18, and S19).

Inter-observer reliability assessment

Before being trained with the ΔHUPV-NC measurement, the agreement, assessed by Cohen’s kappa, between each resident (2 years of experience) and the consensus was 0.29 (95% CI: 0.17, 0.41) and 0.24 (95% CI: 0.1, 0.39). After the training, the inter-rater agreement increased to 0.75 (95% CI: 0.65, 0.85) and 0.72 (95% CI: 0.62, 0.83; Tables 4 and S20–S23). The pre-training diagnostic accuracy of resident-1 and resident-2 were 67% (95% CI: 59, 74) and 66% (95% CI: 58, 73), respectively. After the training, the diagnostic accuracy was improved (resident-1: 81% [95% CI: 74, 87], p = 0.007; resident-2: 81% [95% CI: 74, 87], p = 0.002; Tables 4 and S24–S29).

Table 4 Diagnostic accuracy and inter-observer reliability of two less-experienced residents before and after the training of ΔHUPV-NCa