A deep-learning model for intracranial aneurysm detection on CT angiography images in China: a stepwise, multicentre, early-stage clinical validation study.

BACKGROUND: Artificial intelligence (AI) models in real-world implementation are scarce. Our study aimed to develop a CT angiography (CTA)-based AI model for intracranial aneurysm detection, assess how it helps clinicians improve diagnostic performance, and validate its application in real-world clinical implementation. METHODS: We developed a deep-learning model using 16 546 head and neck CTA examination images from 14 517 patients at eight Chinese hospitals. Using an adapted, stepwise implementation and evaluation, 120 certified clinicians from 15 geographically different hospitals were recruited. Initially, the AI model was externally validated with images of 900 digital subtraction angiography-verified CTA cases (examinations) and compared with the performance of 24 clinicians who each viewed 300 of these cases (stage 1). Next, as a further external validation a multi-reader multi-case study enrolled 48 clinicians to individually review 298 digital subtraction angiography-verified CTA cases (stage 2). The clinicians reviewed each CTA examination twice (ie, with and without the AI model), separated by a 4-week washout period. Then, a randomised open-label comparison study enrolled 48 clinicians to assess the acceptance and performance of this AI model (stage 3). Finally, the model was prospectively deployed and validated in 1562 real-world clinical CTA cases. FINDINGS: The AI model in the internal dataset achieved a patient-level diagnostic sensitivity of 0·957 (95% CI 0·939-0·971) and a higher patient-level diagnostic sensitivity than clinicians (0·943 [0·921-0·961] vs 0·658 [0·644-0·672]; p<0·0001) in the external dataset. In the multi-reader multi-case study, the AI-assisted strategy improved clinicians' diagnostic performance both on a per-patient basis (the area under the receiver operating characteristic curves [AUCs]; 0·795 [0·761-0·830] without AI vs 0·878 [0·850-0·906] with AI; p<0·0001) and a per-aneurysm basis (the area under the weighted alternative free-response receiver operating characteristic curves; 0·765 [0·732-0·799] vs 0·865 [0·839-0·891]; p<0·0001). Reading time decreased with the aid of the AI model (87·5 s vs 82·7 s, p<0·0001). In the randomised open-label comparison study, clinicians in the AI-assisted group had a high acceptance of the AI model (92·6% adoption rate), and a higher AUC when compared with the control group (0·858 [95% CI 0·850-0·866] vs 0·789 [0·780-0·799]; p<0·0001). In the prospective study, the AI model had a 0·51% (8/1570) error rate due to poor-quality CTA images and recognition failure. The model had a high negative predictive value of 0·998 (0·994-1·000) and significantly improved the diagnostic performance of clinicians; AUC improved from 0·787 (95% CI 0·766-0·808) to 0·909 (0·894-0·923; p<0·0001) and patient-level sensitivity improved from 0·590 (0·511-0·666) to 0·825 (0·759-0·880; p<0·0001). INTERPRETATION: This AI model demonstrated strong clinical potential for intracranial aneurysm detection with improved clinician diagnostic performance, high acceptance, and practical implementation in real-world clinical cases. FUNDING: National Natural Science Foundation of China. TRANSLATION: For the Chinese translation of the abstract see Supplementary Materials section.

Peri-lesion regions in differentiating suspicious breast calcification-only lesions specifically on contrast enhanced mammography.

PURPOSE: The explore the added value of peri-calcification regions on contrast-enhanced mammography (CEM) in the differential diagnosis of breast lesions presenting as only calcification on routine mammogram. METHODS: Patients who underwent CEM because of suspicious calcification-only lesions were included. The test set included patients between March 2017 and March 2019, while the validation set was collected between April 2019 and October 2019. The calcifications were automatically detected and grouped by a machine learning-based computer-aided system. In addition to extracting radiomic features on both low-energy (LE) and recombined (RC) images from the calcification areas, the peri-calcification regions, which is generated by extending the annotation margin radially with gradients from 1 mm to 9 mm, were attempted. Machine learning (ML) models were built to classify calcifications into malignant and benign groups. The diagnostic matrices were also evaluated by combing ML models with subjective reading. RESULTS: Models for LE (significant features: wavelet-LLL_glcm_Imc2_MLO; wavelet-HLL_firstorder_Entropy_MLO; wavelet-LHH_glcm_DifferenceVariance_CC; wavelet-HLL_glcm_SumEntropy_MLO;wavelet-HLH_glrlm_ShortRunLowGray LevelEmphasis_MLO; original_firstorder_Entropy_MLO; original_shape_Elongation_MLO) and RC (significant features: wavelet-HLH_glszm_GrayLevelNonUniformityNormalized_MLO; wavelet-LLH_firstorder_10Percentile_CC; original_firstorder_Maximum_MLO; wavelet-HHH_glcm_Autocorrelation_MLO; original_shape_Elongation_MLO; wavelet-LHL_glszm_GrayLevelNonUniformityNormalized_MLO; wavelet-LLH_firstorder_RootMeanSquared_MLO) images were set up with 7 features. Areas under the curve (AUCs) of RC models are significantly better than those of LE models with compact and expanded boundary (RC v.s. LE, compact: 0.81 v.s. 0.73, p <  0.05; expanded: 0.89 v.s. 0.81, p <  0.05) and RC models with 3 mm boundary extension yielded the best performance compared to those with other sizes (AUC = 0.89). Combining with radiologists' reading, the 3mm-boundary RC model achieved a sensitivity of 0.871 and negative predictive value of 0.937 with similar accuracy of 0.843 in predicting malignancy. CONCLUSIONS: The machine learning model integrating intra- and peri-calcification regions on CEM has the potential to aid radiologists' performance in predicting malignancy of suspicious breast calcifications.

Detection of Intracranial Aneurysms Using Multiphase CT Angiography with a Deep Learning Model.

RATIONALE AND OBJECTIVES: Determine the effect of a multiphase fusion deep-learning model with automatic phase selection in detection of intracranial aneurysm (IA) from computed tomography angiography (CTA) images. MATERIALS AND METHODS: CTA images of intracranial arteries from patients at Ningbo First Hospital were retrospectively analyzed. Images were randomly classified as training data, internal validation data, or test data. CTA images from cases examined by digital subtraction angiography (DSA) were examined for independent validation. A deep-learning model was constructed by automatic phase selection of multiphase fusion, and compared to the single-phase algorithm to evaluate algorithm sensitivity. RESULTS: We analyzed 1110 patients (1493 aneurysms) as training data, 139 patients (174 aneurysms) as internal validation data, and 134 patients (175 aneurysms) as test data. The sensitivity of the multiphase analysis of the internal validation data, test data, and independent validation data were greater than from the single-phase analysis. The recall of the multiphase selection was greater or equal to that of single-phase selection in the aneurysm position, shape, size, and rupture status. Use of the test data to determine the presence and absence of aneurysm rupture led to a recall from multiphase selection of 94.8% and 87.6% respectively; both of these values were greater than those from single-phase selection (89.6% and 79.4%). CONCLUSION: A multiphase fusion deep learning model with automatic phase selection provided automated detection of IAs with high sensitivity.

Graph Convolution Based Cross-Network Multiscale Feature Fusion for Deep Vessel Segmentation.

Vessel segmentation is widely used to help with vascular disease diagnosis. Vessels reconstructed using existing methods are often not sufficiently accurate to meet clinical use standards. This is because 3D vessel structures are highly complicated and exhibit unique characteristics, including sparsity and anisotropy. In this paper, we propose a novel hybrid deep neural network for vessel segmentation. Our network consists of two cascaded subnetworks performing initial and refined segmentation respectively. The second subnetwork further has two tightly coupled components, a traditional CNN-based U-Net and a graph U-Net. Cross-network multi-scale feature fusion is performed between these two U-shaped networks to effectively support high-quality vessel segmentation. The entire cascaded network can be trained from end to end. The graph in the second subnetwork is constructed according to a vessel probability map as well as appearance and semantic similarities in the original CT volume. To tackle the challenges caused by the sparsity and anisotropy of vessels, a higher percentage of graph nodes are distributed in areas that potentially contain vessels while a higher percentage of edges follow the orientation of potential nearby vessels. Extensive experiments demonstrate our deep network achieves state-of-the-art 3D vessel segmentation performance on multiple public and in-house datasets.

Evaluation of the peritumoral features using radiomics and deep learning technology in non-spiculated and noncalcified masses of the breast on mammography.

Objective: To assess the significance of peritumoral features based on deep learning in classifying non-spiculated and noncalcified masses (NSNCM) on mammography. Methods: We retrospectively screened the digital mammography data of 2254 patients who underwent surgery for breast lesions in Harbin Medical University Cancer Hospital from January to December 2018. Deep learning and radiomics models were constructed. The classification efficacy in ROI and patient levels of AUC, accuracy, sensitivity, and specificity were compared. Stratified analysis was conducted to analyze the influence of primary factors on the AUC of the deep learning model. The image filter and CAM were used to visualize the radiomics and depth features. Results: For 1298 included patients, 771 (59.4%) were benign, and 527 (40.6%) were malignant. The best model was the deep learning combined model (2 mm), in which the AUC was 0.884 (P < 0.05); especially the AUC of breast composition B reached 0.941. All the deep learning models were superior to the radiomics models (P < 0.05), and the class activation map (CAM) showed a high expression of signals around the tumor of the deep learning model. The deep learning model achieved higher AUC for large size, age >60 years, and breast composition type B (P < 0.05). Conclusion: Combining the tumoral and peritumoral features resulted in better identification of malignant NSNCM on mammography, and the performance of the deep learning model exceeded the radiomics model. Age, tumor size, and the breast composition type are essential for diagnosis.

Prediction of Carotid In-Stent Restenosis by Computed Tomography Angiography Carotid Plaque-Based Radiomics.

In-stent restenosis (ISR) after carotid artery stenting (CAS) critically influences long-term CAS benefits and safety. The study was aimed at screening preoperative ISR-predictive features and developing predictive models. Thus, we retrospectively analyzed clinical and imaging data of 221 patients who underwent pre-CAS carotid computed tomography angiography (CTA) and whose digital subtraction angiography data for verifying ISR presence were available. Carotid plaque characteristics determined using CTA were used to build a traditional model. Backward elimination (likelihood ratio) was used for the radiomics model. Furthermore, a combined model was built using the traditional and radiomics features. Five-fold cross-validation was used to evaluate the accuracy of the trained classifier and stability of the selected features. Follow-up angiography showed ISR in 30 patients. Carotid plaque length and thickness were independently associated with ISR (multivariate analysis); regarding the conventional model, the area under the curve (AUC) was 0.84 and 0.82 in the training and validation cohorts, respectively. The corresponding AUC values for the radiomics-based model were 0.87 and 0.82, and those for the optimal combined model were 0.88 and 0.83. Plaque length and thickness could independently predict post-CAS ISR, and the combination of radiomics and plaque features afforded the best predictive performance.

Radiomics versus Conventional Assessment to Identify Symptomatic Participants at Carotid Computed Tomography Angiography.

INTRODUCTION: Carotid computed tomography angiography (CTA) is routinely used for evaluating the atherosclerotic process. Radiomics allows the extraction of imaging markers of lesion heterogeneity and spatial complexity. These quantitative features can be used as the input for machine learning (ML). Therefore, in this study, we aimed to evaluate the diagnostic performance of radiomics-based ML assessment of carotid CTA data to identify symptomatic patients with carotid artery atherosclerosis. METHODS: In this retrospective study, participants with carotid artery atherosclerosis who underwent carotid CTA and brain magnetic resonance imaging from May 2010 to December 2017 were studied. The participants were grouped into symptomatic and asymptomatic groups according to their recent symptoms (determination of ipsilateral ischemic stroke). Eight conventional plaque features and 2,107 radiomics parameters were extracted from carotid CTA images. A radiomics-based ML model was fitted on the training set, and the radiomics-based ML model and conventional assessment were compared using the area under the curve (AUC) to identify symptomatic participants. RESULTS: After excluding participants with other stroke sources, 120 patients with 148 carotid arteries were analyzed. Of these 148 carotid arteries, 34 (22.97%) were classified into the symptomatic group. Plaque ulceration (odds ratio [OR] = 0.257; 95% confidence interval [CI], 0.094-0.698) and plaque enhancement (OR = 0.305; 95% CI, 0.094-0.988) were associated with the symptomatic status. Twenty radiomics parameters were chosen to be inputs in the radiomics-based ML model. In the identification of symptomatic participants, the discriminatory value of the radiomics-based ML model was significantly higher than that of the conventional assessment (AUC = 0.858 vs. AUC = 0.706, p = 0.021). CONCLUSION: Radiomics-based ML analysis improves the discriminatory power of carotid CTA in the identification of recent ischemic symptoms in patients with carotid artery atherosclerosis.

Act Like a Radiologist: Towards Reliable Multi-View Correspondence Reasoning for Mammogram Mass Detection.

Mammogram mass detection is crucial for diagnosing and preventing the breast cancers in clinical practice. The complementary effect of multi-view mammogram images provides valuable information about the breast anatomical prior structure and is of great significance in digital mammography interpretation. However, unlike radiologists who can utilize the natural reasoning ability to identify masses based on multiple mammographic views, how to endow the existing object detection models with the capability of multi-view reasoning is vital for decision-making in clinical diagnosis but remains the boundary to explore. In this paper, we propose an anatomy-aware graph convolutional network (AGN), which is tailored for mammogram mass detection and endows existing detection methods with multi-view reasoning ability. The proposed AGN consists of three steps. First, we introduce a bipartite graph convolutional network (BGN) to model the intrinsic geometric and semantic relations of ipsilateral views. Second, considering that the visual asymmetry of bilateral views is widely adopted in clinical practice to assist the diagnosis of breast lesions, we propose an inception graph convolutional network (IGN) to model the structural similarities of bilateral views. Finally, based on the constructed graphs, the multi-view information is propagated through nodes methodically, which equips the features learned from the examined view with multi-view reasoning ability. Experiments on two standard benchmarks reveal that AGN significantly exceeds the state-of-the-art performance. Visualization results show that AGN provides interpretable visual cues for clinical diagnosis.

Bilateral Asymmetry Guided Counterfactual Generating Network for Mammogram Classification.

Mammogram benign or malignant classification with only image-level labels is challenging due to the absence of lesion annotations. Motivated by the symmetric prior that the lesions on one side of breasts rarely appear in the corresponding areas on the other side, we explore to answer a counterfactual question to identify the lesion areas. This counterfactual question means: given an image with lesions, how would the features have behaved if there were no lesions in the image? To answer this question, we derive a new theoretical result based on the symmetric prior. Specifically, by building a causal model that entails such a prior for bilateral images, we identify to optimize the distances in distribution between i) the counterfactual features and the target side's features in lesion-free areas; and ii) the counterfactual features and the reference side's features in lesion areas. To realize these optimizations for better benign/malignant classification, we propose a counterfactual generative network, which is mainly composed of Generator Adversarial Network and a prediction feedback mechanism, they are optimized jointly and prompt each other. Specifically, the former can further improve the classi?cation performance by generating counterfactual features to calculate lesion areas. On the other hand, the latter helps counterfactual generation by the supervision of classification loss. The utility of our method and the effectiveness of each module in our model can be verified by state-of-the-art performance on INBreast and an in-house dataset and ablation studies.

Compare and contrast: Detecting mammographic soft-tissue lesions with C2-Net.

Detecting breast soft-tissue lesions including masses, structural distortions and asymmetries is of great importance due to the high risk leading to breast cancer. Most existing deep learning based approaches detect lesions with only unilateral images. However, multi-view mammogram images provide highly related and complementary information which helps to make the clinical analysis more comprehensive and reliable. In this paper, we propose a multi-view network for breast soft-tissue lesion detection called C2-Net (Compare and Contrast, C2) that fuses information across different views. The proposed model contains the following three modules. The spatial context enhancing (SCE) module compares ipsilateral views and extracts complementary features to model lesion inherent 3D structure. The multi-scale kernel pooling (MKP) module contrasts contralateral views with added misalignment tolerance. Finally, the logic guided fusion (LGF) module fuses multi-view features by enhancing logic modeling capacity. Experimental results on both the public DDSM dataset and the in-house multi-center dataset demonstrate that the proposed method has achieved state-of-the-art performance.

Improving the Diagnostic Accuracy of Breast BI-RADS 4 Microcalcification-Only Lesions Using Contrast-Enhanced Mammography.

BACKGROUND: Contrast-enhanced mammography (CEM) is a novel breast imaging technique that can provide additional information of breast tissue blood supply. This study aimed to test the possibility of CEM in improving the diagnostic accuracy of Breast Imaging Reporting and Data System (BI-RADS) 4 calcification-only lesions with consideration of morphology and distribution. PATIENTS AND METHODS: Data of patients with suspicious malignant calcification-only lesions (BI-RADS 4) on low-energy CEM and proved pathologic diagnoses were retrospectively collected. Two junior radiologists independently reviewed the two sets of CEM images, low-energy images (LE) to describe the calcifications by morphology and distribution type, and recombined images (CE) to record the presence of enhancement. Low-risk and high-risk groups were divided by calcification morphology, distribution, and both, respectively. Positive predictive values and misdiagnosis rates (MDR) were compared between LE-only reading and CE reading. Diagnostic performance was also tested using machine learning method. RESULTS: The study included 74 lesions (26 malignant and 48 benign). Positive predictive values were significantly higher and MDRs were significantly lower using CE images than using LE alone for both the low-risk morphology type and low-risk distribution type (P < .05). MDRs were significantly lower when using CE images (18.18%-24.00%) than using LE images alone in low-risk group (76.36%-80.00%) (P < .05). Using a machine learning method, significant improvements in the area under the receiver operating characteristic curve were observed in both low-risk and high-risk groups. CONCLUSION: CEM has the potential to aid in the diagnosis of BI-RADS 4 calcification-only lesions; in particular, those presented as low risk in morphology and/or distribution may benefit more.

[Performance of Deep-learning-based Artificial Intelligence on Detection of Pulmonary Nodules in Chest CT].

BACKGROUND: The detection of pulmonary nodules is a key step to achieving the early diagnosis and therapy of lung cancer. Deep learning based Artificial intelligence (AI) presents as the state of the art in the area of nodule detection, however, a validation with clinical data is necessary for further application. Therefore, the aim of this study is to evaluate the performance of AI in the detection of malignant and non-calcified nodules in chest CT. METHODS: Two hundred chest computed tomography (CT) data were randomly selected from a self-built nodule database from Tianjin Medical University General Hospital. Both the pathology confirmed lung cancers and the nodules in the process of follow-up were included. All CTs were processed by AI and the results were compared with that of radiologists retrieved from the original medical reports. The ground truths were further determined by two experienced radiologists. The size and characteristics of the nodules were evaluated as well. The sensitivity and false positive rate were used to evaluate the effectiveness of AI and radiologists in detecting nodules. The McNemar test was used to determine whether there was a significant difference. RESULTS: A total of 889 non-calcified nodules were determined by experts on chest CT, including 133 lung cancers. Of them, 442 nodules were less than 5 mm. The cancer detection rates of AI and radiologists are 100%. The sensitivity of AI on nodule detection was significantly higher than that of radiologists (99.1% vs 43%, P<0.001). The false-positive rate of AI was 4.9 per CT and decreased to 1.5 when nodules less than 5 mm were excluded. CONCLUSIONS: AI achieves the detection of all malignancies and improve the sensitivity of pulmonary nodules detection beyond radiologists, with a low false positive rate after excluding small nodules.