The uncovered biases and errors in clinical determination of bone age by using deep learning models.

OBJECTIVES: To evaluate AI biases and errors in estimating bone age (BA) by comparing AI and radiologists' clinical determinations of BA. METHODS: We established three deep learning models from a Chinese private dataset (CHNm), an American public dataset (USAm), and a joint dataset combining the above two datasets (JOIm). The test data CHNt (n = 1246) were labeled by ten senior pediatric radiologists. The effects of data site differences, interpretation bias, and interobserver variability on BA assessment were evaluated. The differences between the AI models' and radiologists' clinical determinations of BA (normal, advanced, and delayed BA groups by using the Brush data) were evaluated by the chi-square test and Kappa values. The heatmaps of CHNm-CHNt were generated by using Grad-CAM. RESULTS: We obtained an MAD value of 0.42 years on CHNm-CHNt; this result indicated an appropriate accuracy for the whole group but did not indicate an accurate estimation of individual BA because with a kappa value of 0.714, the agreement between AI and human clinical determinations of BA was significantly different. The features of the heatmaps were not fully consistent with the human vision on the X-ray films. Variable performance in BA estimation by different AI models and the disagreement between AI and radiologists' clinical determinations of BA may be caused by data biases, including patients' sex and age, institutions, and radiologists. CONCLUSIONS: The deep learning models outperform external validation in predicting BA on both internal and joint datasets. However, the biases and errors in the models' clinical determinations of child development should be carefully considered. KEY POINTS: • With a kappa value of 0.714, clinical determinations of bone age by using AI did not accord well with clinical determinations by radiologists. • Several biases, including patients' sex and age, institutions, and radiologists, may cause variable performance by AI bone age models and disagreement between AI and radiologists' clinical determinations of bone age. • AI heatmaps of bone age were not fully consistent with human vision on X-ray films.

Diagnostic Value of Multimodal Magnetic Resonance Imaging in Discriminating Between Metastatic and Non-Metastatic Pelvic Lymph Nodes in Cervical Cancer.

Background: The status of pelvic lymph node (PLN) metastasis affects treatment and prognosis plans in patients with cervical cancer. However, it is hard to be diagnosed in clinical practice. Purpose: The present study aimed to evaluate the diagnostic value of multimodal magnetic resonance imaging (MRI) in discriminating between metastatic and non-metastatic pelvic lymph nodes (PLNs) in cervical cancer. Methods: This retrospective study analyzed MRIs of 209 PLNs in 25 women with pathologically proven cervical cancer. All PLNs had been assessed by pre-treatment multimodal MRIs, and their status was finally confirmed by histopathology. In conventional MRI, lymph node characteristics were compared between metastatic and non-metastatic PLNs. Signal intensity, time-intensity curve (TIC) patterns minimal and mean apparent diffusion coefficients (ADC) were compared between them in DWI. In DCE-MRI, quantitative (Ktrans, Kep and Ve) analyses were performed on DCE-MRI sequences, and their predictive values were analyzed by ROC curves. Results: Of 209 PLNs, 22 (10.53%) were metastases and 187 (89.47%) were non-metastases at histopathologic examination. Considering a comparison of lymph node characteristics, the short axis size, the long axis size, and the boundary differed significantly between the two groups (P<0.05).The differences in ADCmin, TIC types, Ktrans and Ve between metastatic and non-metastatic PLNs were significant as well (P<0.05). The good diagnostic performance of multimodal MRI was shown in discriminating between metastatic and non-metastatic PLNs, with the sensitivity of 85.0% (17/20), specificity of 97.3% (184/189), and accuracy of 96.2% (201/209). ROC analyses showed that the diagnostic accuracy of ADCmin, Ktrans and Ve for discriminating between metastatic and non-metastatic PLNs in cervical cancer was 83.7%, 91.4%, and 92.4% with the cut-off values of 0.72 × 10-3mm2/s, 0.52 min-1, and 0.53 min-1, respectively. Conclusion: Multimodal MRI showed good diagnostic performance in determining PLN status in cervical cancer.

A dual-path neural network fusing dual-sequence magnetic resonance image features for detection of placenta accrete spectrum (PAS) disorder.

With the increase of various risk factors such as cesarean section and abortion, placenta accrete spectrum (PAS) disorder is happening more frequently year by year. Therefore, prenatal prediction of PAS is of crucial practical significance. Magnetic resonance imaging (MRI) quality will not be affected by fetal position, maternal size, amniotic fluid volume, etc., which has gradually become an important means for prenatal diagnosis of PAS. In clinical practice, T2-weighted imaging (T2WI) magnetic resonance (MR) images are used to reflect the placental signal and T1-weighted imaging (T1WI) MR images are used to reflect bleeding, both plays a key role in the diagnosis of PAS. However, it is difficult for traditional MR image analysis methods to extract multi-sequence MR image features simultaneously and assign corresponding weights to predict PAS according to their importance. To address this problem, we propose a dual-path neural network fused with a multi-head attention module to detect PAS. The model first uses a dual-path neural network to extract T2WI and T1WI MR image features separately, and then combines these features. The multi-head attention module learns multiple different attention weights to focus on different aspects of the placental image to generate highly discriminative final features. The experimental results on the dataset we constructed demonstrate a superior performance of the proposed method over state-of-the-art techniques in prenatal diagnosis of PAS. Specifically, the model we trained achieves 88.6% accuracy and 89.9% F1-score on the independent validation set, which shows a clear advantage over methods that only use a single sequence of MR images.

The diagnostic value of MRI for architectural distortion categorized as BI-RADS category 3-4 by mammography.

BACKGROUND: Architectural distortion is a common mammographic sign that can be benign or malignant. This study investigated the diagnostic value of magnetic resonance imaging (MRI) for architectural distortions that were category 3-4 under the breast imaging reporting and data system (BI-RADS) by mammography. METHODS: We retrospectively analyzed 219 pathologically confirmed lesions in 208 patients who had BI-RADS category 3-4 architectural distortion in mammography images. Two radiologists described and categorized the architectural distortion and assigned the BI-RADS categories to the corresponding lesions on MRI images. Using the postoperative pathological diagnosis as the gold standard, we performed receiver operating characteristic (ROC) analysis for the efficacy of mammography and MRI in differentiating patients with benign or malignant lesions. RESULTS: Totally 151 benign lesions and 68 malignant lesions were confirmed. According to the full-field digital mammography (FFDM), 82 lesions were in BI-RADS category 3, 104 lesions in 4A, 29 lesions in 4B, and 4 lesions in 4C. The positive predictive values of FFDM for BI-RADS categories 3, 4A, 4B, and 4C were 13.4% (11/82), 27.9% (29/104), 82.8% (24/29), and 100.0% (4/4), respectively. According to MRI, 59 lesions were in BI-RADS categories 1-2, 87 lesions in 3, 39 lesions in 4, and 34 lesions in 5, with their positive predictive values being 0.0% (0/58), 2.3% (2/87), 89.7% (35/39), and 100.0% (34/34), respectively. The area under the ROC curve (AUC) of breast benign and malignant lesions differentiated by FFDM was 0.647, and the diagnostic sensitivity, specificity, and Youden index were 86.3%, 41.7%, and 0.280, respectively. The AUC of FFDM combined with dynamic contrast-enhanced MRI (DCE-MRI) in differentiating breast benign vs. malignant lesions was 0.851, and the diagnostic sensitivity, specificity, and Youden index were 89.2%, 80.7%, and 0.699, respectively. The AUC of FFDM combined with DCE-MRI and the apparent diffusion coefficient (ADC) in differentiating benign vs. malignant lesions was 0.983, and the diagnostic sensitivity, specificity, and Youden index were 98.1%, 97.5%, and 0.956, respectively. CONCLUSIONS: MRI can improve the diagnostic efficiency of mammography in diagnosing BI-RADS category 3-4 architectural distortions and can help in the qualitative diagnosis of architectural distortion lesions.

[Imaging findings of bone infarction in children with systemic lupus erythematosus].

OBJECTIVE: To investigate diagnostic value of MRI, X ray and CT for bone infarction in children with systemic lupus erythematosus. METHODS: Eleven systemic lupus erythematosus children with bone infarction were retrospectively analyzed from January 2015 to January 2017 , and tested by MRI, X-ray and CT. Among them, including 1 male and 10 females aged from 6 to 16 years old with an average of 13 years old. All patients were detected by MRI, 9 patients were detected by X-ray and 3 patients were detected by CT, imaging findings were analyzed. RESULTS: The location of bone infarction involved 60 sits, 30 sites located on metaphyseal-diaphyseal region, 8 located on patella, 21 located on epiphysis, and 1 located on talus. Focus of 11 patients were detected by MRI, the main manifestation showed geographic change, long T1 and T2 signal could seen around focus, and showed double ring sign and three ring sign; 5 of 9 patients by X-ray examination detected focus;2 of 3 patients by CT examination detected focus. No abnormity seen at early stage by X-ray and CT examination, and low density focus around harden edge at chronic stage. CONCLUSIONS: MRI could display bone fracture at early stage, X-ray and CT could only display lesion at chronic stage, MRI is the most effective method in diagnosing bone infarction.