Bone Age Assessment Using Artificial Intelligence in Korean Pediatric Population: A Comparison of Deep-Learning Models Trained With Healthy Chronological and Greulich-Pyle Ages as Labels.

OBJECTIVE: To develop a deep-learning-based bone age prediction model optimized for Korean children and adolescents and evaluate its feasibility by comparing it with a Greulich-Pyle-based deep-learning model. MATERIALS AND METHODS: A convolutional neural network was trained to predict age according to the bone development shown on a hand radiograph (bone age) using 21036 hand radiographs of Korean children and adolescents without known bone development-affecting diseases/conditions obtained between 1998 and 2019 (median age [interquartile range {IQR}], 9 [7-12] years; male:female, 11794:9242) and their chronological ages as labels (Korean model). We constructed 2 separate external datasets consisting of Korean children and adolescents with healthy bone development (Institution 1: n = 343; median age [IQR], 10 [4-15] years; male: female, 183:160; Institution 2: n = 321; median age [IQR], 9 [5-14] years; male: female, 164:157) to test the model performance. The mean absolute error (MAE), root mean square error (RMSE), and proportions of bone age predictions within 6, 12, 18, and 24 months of the reference age (chronological age) were compared between the Korean model and a commercial model (VUNO Med-BoneAge version 1.1; VUNO) trained with Greulich-Pyle-based age as the label (GP-based model). RESULTS: Compared with the GP-based model, the Korean model showed a lower RMSE (11.2 vs. 13.8 months; P = 0.004) and MAE (8.2 vs. 10.5 months; P = 0.002), a higher proportion of bone age predictions within 18 months of chronological age (88.3% vs. 82.2%; P = 0.031) for Institution 1, and a lower MAE (9.5 vs. 11.0 months; P = 0.022) and higher proportion of bone age predictions within 6 months (44.5% vs. 36.4%; P = 0.044) for Institution 2. CONCLUSION: The Korean model trained using the chronological ages of Korean children and adolescents without known bone development-affecting diseases/conditions as labels performed better in bone age assessment than the GP-based model in the Korean pediatric population. Further validation is required to confirm its accuracy.

Diagnostic performance of the 2021 Korean thyroid imaging reporting and data system in pediatric thyroid nodules.

OBJECTIVES: To evaluate the diagnostic performance of 2021 K-TIRADS biopsy criteria for detecting malignant thyroid nodules in a pediatric population, making comparisons with 2016 K-TIRADS. METHODS: This retrospective study included pediatric patients with histopathologically confirmed diagnoses. The diagnostic performance of 2021 K-TIRADS was compared with that of 2016 K-TIRADS. Simulation studies were performed by changing biopsy cut-off sizes for K-TIRADS 5 to 1.0 cm (K-TIRADS5-1.0cm) and 0.5 cm (K-TIRADS5-0.5cm), and for K-TIRADS 4 to 1.0 cm (K-TIRADS4-1.0cm) and 1.0-1.5 cm (K-TIRADS4-1.0~1.5cm). Subgroup analysis was performed in small (< 1.5 cm) and large nodules (≥ 1.5 cm). RESULTS: Two hundred seventy-seven thyroid nodules (54.9% malignant) from 221 pediatric patients were analyzed. All simulated 2021 K-TIRADS showed higher accuracy than 2016 K-TIRADS. Compared with 2021 K-TIRADS5-1.0cm, 2021 K-TIRADS5-0.5cm showed lower specificity (51.6% vs. 47.9%; p = 0.004) but higher sensitivity (77.2% vs. 90.3%; p < 0.001) and accuracy (62.7% vs. 68.9%; p < 0.001). Compared with 2021 K-TIRADS4-1.0cm, 2021 K-TIRADS4-1.0~1.5cm showed higher specificity (44.9% vs. 47.9%; p = 0.018) without significant difference in other diagnostic measures. Compared with 2016 K-TIRADS, 2021 K-TIRADS (biopsy cut-offs, 0.5 cm for K-TIRADS 5; 1.0-1.5 cm for K-TIRADS 4) showed higher sensitivity (34.0% vs. 67.3%; p < 0.001) while maintaining specificity (89.4% vs. 88.2%; p = 0.790) in small nodules, and higher specificity (5.9% vs. 25.4%; p < 0.001) while maintaining sensitivity (100% vs. 98.7%; p = 0.132) in large nodules. CONCLUSIONS: In pediatric patients, 2021 K-TIRADS showed superior diagnostic accuracy to 2016 K-TIRADS, especially with a biopsy cut-off of 0.5 cm for K-TIRADS 5 and 1.0-1.5 cm for K-TIRADS 4. KEY POINTS: • All simulated 2021 K-TIRADS showed higher accuracy than 2016 K-TIRADS. • 2021 K-TIRADS with cut-off size for K-TIRADS 5 of 0.5 cm showed lower specificity but higher sensitivity and accuracy than that of 1.0 cm. • Compared with 2016 K-TIRADS, 2021 K-TIRADS (biopsy cut-offs, 0.5 cm for K-TIRADS 5; 1.0-1.5 cm for K-TIRADS 4) showed higher sensitivity while maintaining specificity in small nodules, and higher specificity while maintaining sensitivity in large nodules.

Diagnostic Performance of Five Adult-based US Risk Stratification Systems in Pediatric Thyroid Nodules.

Background The validation of adult-based US risk stratification systems (RSSs) in the discrimination of malignant thyroid nodules in a pediatric population remains lacking. Purpose To estimate and compare the diagnostic performance of pediatric US RSSs based on five adult-based RSSs in the discrimination of malignant thyroid nodules in a pediatric sample. Materials and methods Pediatric patients (age ≤18 years) with histopathologically confirmed US-detected thyroid nodules at a tertiary referral hospital between January 2000 and April 2020 were analyzed retrospectively. The diagnostic performance of US-based fine-needle aspiration biopsy (FNAB) criteria in thyroid cancer detection was estimated. The following sensitivity analyses were performed: (a) scenario 1: nodules smaller than 1 cm, with the highest category additionally biopsied; (b) scenario 2, application of American College of Radiology Thyroid Imaging Reporting and Data System nodule size cutoffs to other RSSs; (c) scenario 3, scenarios 1 and 2 together. Generalized estimating equations (GEEs) were used for estimation. Results A total of 277 thyroid nodules in 221 pediatric patients (median age, 16 years [interquartile range {IQR}, 13-17]; 172 female; 152 of 277 patients [55%] malignant) were analyzed. The GEE-estimated sensitivity and specificity ranged from 70% to 78% (104 to 119 of 152 patients, based on each reader's interpretation) and from 42% to 78% (49 of 124 patients to 103 of 125 patients). In scenario 1, the missed malignancy rate was reduced from 32%-38% (41 of 134 patients to 34 of 83 patients) to 15%-21% (eight of 59 patients to 28 of 127 patients). In scenario 2, the unnecessary biopsy rate was reduced from 35%-39% (60 of 176 patients to 68 of 175 patients) to 20%-34% (18 of 109 patients to 62 of 179 patients). The highest accuracy was noted in scenario 3 (range, 71%-81%; 199 of 277 patients to 216 of 262 patients). Conclusion The diagnostic performances of the fine-needle aspiration biopsy criteria of five adult-based risk stratification systems were acceptable in the pediatric population and were improved by applying the American College of Radiology Thyroid Imaging Reporting and Data System size cutoff for nodules 1 cm or larger and allowing biopsy of the highest category nodules smaller than 1 cm. © RSNA, 2022 Online supplemental material is available for this article.

Re-Assessment of Applicability of Greulich and Pyle-Based Bone Age to Korean Children Using Manual and Deep Learning-Based Automated Method.

PURPOSE: To evaluate the applicability of Greulich-Pyle (GP) standards to bone age (BA) assessment in healthy Korean children using manual and deep learning-based methods. MATERIALS AND METHODS: We collected 485 hand radiographs of healthy children aged 2-17 years (262 boys) between 2008 and 2017. Based on GP method, BA was assessed manually by two radiologists and automatically by two deep learning-based BA assessment (DLBAA), which estimated GP-assigned (original model) and optimal (modified model) BAs. Estimated BA was compared to chronological age (CA) using intraclass correlation (ICC), Bland-Altman analysis, linear regression, mean absolute error, and root mean square error. The proportion of children showing a difference >12 months between the estimated BA and CA was calculated. RESULTS: CA and all estimated BA showed excellent agreement (ICC ≥0.978, p<0.001) and significant positive linear correlations (R²≥0.935, p<0.001). The estimated BA of all methods showed systematic bias and tended to be lower than CA in younger patients, and higher than CA in older patients (regression slopes ≤-0.11, p<0.001). The mean absolute error of radiologist 1, radiologist 2, original, and modified DLBAA models were 13.09, 13.12, 11.52, and 11.31 months, respectively. The difference between estimated BA and CA was >12 months in 44.3%, 44.5%, 39.2%, and 36.1% for radiologist 1, radiologist 2, original, and modified DLBAA models, respectively. CONCLUSION: Contemporary healthy Korean children showed different rates of skeletal development than GP standard-BA, and systemic bias should be considered when determining children's skeletal maturation.

Accuracy of attenuation imaging in the assessment of pediatric hepatic steatosis: correlation with the controlled attenuation parameter.

PURPOSE: This study evaluated the accuracy of attenuation imaging (ATI) for the assessment of hepatic steatosis in pediatric patients, in comparison with the FibroScan vibration-controlled transient elastography controlled attenuation parameter (CAP). METHODS: Consecutive pediatric patients referred for evaluation of obesity who underwent both ATI and FibroScan between February 2020 and September 2021 were included. The correlation between attenuation coefficient (AC) and CAP values was assessed using the Spearman test. The AC cutoff value for discriminating hepatic steatosis corresponding to a CAP value of 241 dB/m was calculated. Multivariable linear regression analysis was performed to estimate the strength of the association between AC and CAP. The diagnostic accuracy of AC cutoffs was estimated using the imperfect gold-standard methodology based on a two-level Bayesian latent class model. RESULTS: Seventy patients (median age, 12.5 years; interquartile range, 11.0 to 14.0 years; male:female, 58:12) were included. AC and CAP showed a moderate-to-good correlation (ρ =0.646, P<0.001). Multivariable regression analysis affirmed the significant association between AC and CAP (P<0.001). The correlation was not evident in patients with a body mass index ≥30 kg/m2 (ρ=-0.202, P=0.551). Linear regression revealed that an AC cutoff of 0.66 dB/cm/MHz corresponded to a CAP of 241 dB/m (sensitivity, 0.93; 95% confidence interval [CI], 0.85 to 0.98 and specificity, 0.87; 95% CI, 0.56 to 1.00). CONCLUSION: ATI showed an acceptable correlation with CAP values in a pediatric population, especially in patients with a body mass index <30 kg/m2. An AC cutoff of 0.66 dB/cm/MHz, corresponding to a CAP of 241 dB/m, can accurately diagnose hepatic steatosis.