Artificial intelligence assistance improves the accuracy and efficiency of intracranial aneurysm detection with CT angiography.

PURPOSE: The aim of this study was to evaluate whether a novel head and neck artificial intelligence (AI)-assisted diagnostic system based on a three-dimensional convolutional neural network (3D-CNN) could improve the accuracy, efficiency and working mode of intracranial aneurysm (IA) detection. METHODS: A total of 212 patients who underwent computed tomography angiography (CTA) and digital subtraction angiography (DSA) were retrospectively included. We used three diagnostic modes to detect IAs with CTA: AI, physicians and AI + physicians. Taking the diagnostic results of DSA as the gold standard, the sensitivity, specificity, accuracy, mean reporting time, and interobserver consistency of the three diagnostic modes were calculated and compared at the patient and lesion levels. RESULTS: Of 212 patients, 179 were diagnosed with IAs by DSA, and 224 IAs were analyzed. The sensitivity, specificity and accuracy of the AI system in diagnosing aneurysms were 84.9% (95% confidence interval [CI], 78.9-89.5%), 18.2% (95% CI, 8.2-34.8%) and 74.5% (95% CI, 68.3-80.0%) at the patient-level, and 77.2% (95% CI, 71.3-82.3%), 14.0% (95% CI, 6.2-27.6%) and 67.0% (95% CI, 61.2-72.4%) at the lesion-level, respectively. With AI assistance, junior physicians had the similar diagnostic performance as senior physicians at the patient (sensitivity 95.0% vs. 95.0%, specificity 48.5% vs. 57.6%, accuracy 87.7% vs. 89.2%, p = 0.690) and lesion levels (sensitivity 88.0% vs. 89.7%, specificity 32.0% vs. 38.0%, accuracy 77.8% vs. 80.3%, p = 1.000), especially for aneurysms < 5 mm (sensitivity 83.2% vs. 87.6%, specificity 60.0% vs. 63.2%, accuracy 75.4% vs. 78.9%, p = 0.424). The reporting efficiency of junior and senior physicians improved by 20.7% (141.1 ± 52.6 s to 111.9 ± 46.3 s, p = 0.004) and 18.8% (113.2 ± 42.5 s to 91.9 ± 41.2 s, p = 0.011), respectively. CONCLUSIONS: This 3D-CNN-based AI system significantly improved the accuracy and efficiency of physician detection of IA. The AI + physicians work mode could have a major influence on daily clinical practice and clinical research.

A deep-learning model for identifying fresh vertebral compression fractures on digital radiography.

OBJECTIVES: To develop a deep-learning (DL) model for identifying fresh VCFs from digital radiography (DR), with magnetic resonance imaging (MRI) as the reference standard. METHODS: Patients with lumbar VCFs were retrospectively enrolled from January 2011 to May 2020. All patients underwent DR and MRI scanning. VCFs were categorized as fresh or old according to MRI results, and the VCF grade and type were assessed. The raw DR data were sent to InferScholar Center for annotation. A DL-based prediction model was built, and its diagnostic performance was evaluated. The DeLong test was applied to assess differences in ROC curves between different models. RESULTS: A total of 1877 VCFs in 1099 patients were included in our study and randomly divided into development (n = 824 patients) and test (n = 275 patients) datasets. The ensemble model identified fresh and old VCFs, reaching an AUC of 0.80 (95% confidence interval [CI], 0.77-0.83), an accuracy of 74% (95% CI, 72-77%), a sensitivity of 80% (95% CI, 77-83%), and a specificity of 68% (95% CI, 63-72%). Lateral (AUC, 0.83) views exhibited better performance than anteroposterior views (AUC, 0.77), and the best performance among respective subgroupings was obtained for grade 3 (AUC, 0.89) and crush-type (AUC, 0.87) subgroups. CONCLUSION: The proposed DL model achieved adequate performance in identifying fresh VCFs from DR. KEY POINTS: • The ensemble deep-learning model identified fresh VCFs from DR, reaching an AUC of 0.80, an accuracy of 74%, a sensitivity of 80%, and a specificity of 68% with the reference standard of MRI. • The lateral views (AUC, 0.83) exhibited better performance than anteroposterior views (AUC, 0.77). • The grade 3 (AUC, 0.89) and crush-type (AUC, 0.87) subgroups showed the best performance among their respective subgroupings.