Automatic radiomic feature extraction using deep learning for angiographic parametric imaging of intracranial aneurysms

Classification, detection, and segmentation performance of image-based AI in intracranial aneurysm: a systematic review

BACKGROUND

The detection and management of intracranial aneurysms (IAs) are vital to prevent life-threatening complications like subarachnoid hemorrhage (SAH). Artificial Intelligence (AI) can analyze medical images, like CTA or MRA, spotting nuances possibly overlooked by humans. Early detection facilitates timely interventions and improved outcomes. Moreover, AI algorithms offer quantitative data on aneurysm attributes, aiding in long-term monitoring and assessing rupture risks.

METHODS

We screened four databases (PubMed, Web of Science, IEEE and Scopus) for studies using artificial intelligence algorithms to identify IA. Based on algorithmic methodologies, we categorized them into classification, segmentation, detection and combined, and then their merits and shortcomings are compared. Subsequently, we elucidate potential challenges that contemporary algorithms might encounter within real-world clinical diagnostic contexts. Then we outline prospective research trajectories and underscore key concerns in this evolving field.

RESULTS

Forty-seven studies of IA recognition based on AI were included based on search and screening criteria. The retrospective results represent that current studies can identify IA in different modal images and predict their risk of rupture and blockage. In clinical diagnosis, AI can effectively improve the diagnostic accuracy of IA and reduce missed detection and false positives.

CONCLUSIONS

The AI algorithm can detect unobtrusive IA more accurately in communicating arteries and cavernous sinus arteries to avoid further expansion. In addition, analyzing aneurysm rupture and blockage before and after surgery can help doctors plan treatment and reduce the uncertainties in the treatment process.

Advances in research and application of artificial intelligence and radiomic predictive models based on intracranial aneurysm images

Intracranial aneurysm is a high-risk disease, with imaging playing a crucial role in their diagnosis and treatment. The rapid advancement of artificial intelligence in imaging technology holds promise for the development of AI-based radiomics predictive models. These models could potentially enable the automatic detection and diagnosis of intracranial aneurysms, assess their status, and predict outcomes, thereby assisting in the creation of personalized treatment plans. In addition, these techniques could improve diagnostic efficiency for physicians and patient prognoses. This article aims to review the progress of artificial intelligence radiomics in the study of intracranial aneurysms, addressing the challenges faced and future prospects, in hopes of introducing new ideas for the precise diagnosis and treatment of intracranial aneurysms.

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.

FSTIF-UNet: A Deep Learning-Based Method Towards Automatic Segmentation of Intracranial Aneurysms in Un-Reconstructed 3D-RA

Segmentation of intracranial aneurysms (IAs) is an important step for the diagnosis and treatment of IAs. However, the process by which clinicians manually recognize and localize IAs is overly labor intensive. This study aims to develop a deep-learning-based framework (defined as FSTIF-UNet) towards IAs segmentation in un-reconstructed 3D Rotational Angiography (3D-RA) images. 3D-RA sequences from 300 patients with IAs from Beijing Tiantan Hospital are enrolled. Inspired by radiologists' clincial skills, a Skip-Review attention mechanism is proposed to repeatedly fuse the long-term spatiotemporal features of several images with the most obvious IA's features (sellected by a pre-detection network). Then, a Conv-LSTM is used to fuse the short-term spatiotemporal features of the selected 15 3D-RA images from the equally-spaced viewing angles. The combination of the two modules realizes the full-scale spatiotemporal information fusion of the 3D-RA sequence. FSTIF-UNet achieves DSC, IoU, Sens, Haus, and F1-Score of 0.9109, 0.8586, 0.9314, 1.358 and 0.8883, respectively, and time taken for network segmentation is 0.89 s/case. The results show significant improvement in IA segmentation performance with FSTIF-UNet compared with baseline networks (with DSC from 0.8486 - 0.8794). The proposed FSTIF-UNet establishes a practical method to assist the radiologists in clinical diagnosis.

Radiomics Nomogram Model Based on TOF-MRA Images: A New Effective Method for Predicting Microaneurysms

Objective

To develop a radiomics nomogram model based on time-of-flight magnetic resonance angiography (TOF-MRA) images for preoperative prediction of true microaneurysms.

Methods

118 patients with Intracranial Aneurysm Sac (40 positive and 78 negative) were enrolled and allocated to training and validation groups (8:2 ratio). Findings of clinical characteristics and MRA features were analyzed. A radiomics signature was built on the basis of reproducible features by using the least absolute shrinkage and selection operator (LASSO) regression algorithm in the training group. The radiomics nomogram model was constructed by combining clinical risk factors and radiomics signature. In order to compare the classification performance of clinical models, radiomics model and radiomics nomogram model, AUC was used to evaluate them. The performance of the radiomics nomogram model was evaluated by calibration curve and decision curve analysis.

Results

Eleven features were selected to develop radiomics model with AUC of 0.875 (95% CI 0.78-0.97), sensitivity of 0.84, and specificity of 0.68. The radiomics model achieved a better diagnostic performance than the clinic model (AUC = 0.75, 95% CI: 0.53-0.97) and even radiologists. The radiomics nomogram model, which combines radiomics signature and clinical risk factors, is effective too (AUC = 0.913, 95% CI: 0.87-0.96). Furthermore, the decision curve analysis demonstrated significantly better net benefit in the radiomics nomogram model.

Conclusion

Radiomics features derived from TOF-MRA can reliably be used to build a radiomics nomogram model for effectively differentiating between pseudo microaneurysms and true microaneurysms, and it can provide an objective basis for the selection of clinical treatment plans.

A Review of Artificial Intelligence in Cerebrovascular Disease Imaging: Applications and Challenges

BACKGROUND

A variety of emerging medical imaging technologies based on artificial intelligence have been widely applied in many diseases, but they are still limitedly used in the cerebrovascular field even though the diseases can lead to catastrophic consequences.

OBJECTIVE

This work aims to discuss the current challenges and future directions of artificial intelligence technology in cerebrovascular diseases through reviewing the existing literature related to applications in terms of computer-aided detection, prediction and treatment of cerebrovascular diseases.

METHODS

Based on artificial intelligence applications in four representative cerebrovascular diseases including intracranial aneurysm, arteriovenous malformation, arteriosclerosis and moyamoya disease, this paper systematically reviews studies published between 2006 and 2021 in five databases: National Center for Biotechnology Information, Elsevier Science Direct, IEEE Xplore Digital Library, Web of Science and Springer Link. And three refinement steps were further conducted after identifying relevant literature from these databases.

RESULTS

For the popular research topic, most of the included publications involved computer-aided detection and prediction of aneurysms, while studies about arteriovenous malformation, arteriosclerosis and moyamoya disease showed an upward trend in recent years. Both conventional machine learning and deep learning algorithms were utilized in these publications, but machine learning techniques accounted for a larger proportion.

CONCLUSION

Algorithms related to artificial intelligence, especially deep learning, are promising tools for medical imaging analysis and will enhance the performance of computer-aided detection, prediction and treatment of cerebrovascular diseases.

InterNet: Detection of Active Abdominal Arterial Bleeding Using Emergency Digital Subtraction Angiography Imaging With Two-Stage Deep Learning

Objective

Active abdominal arterial bleeding is an emergency medical condition. Herein, we present our use of this two-stage InterNet model for detection of active abdominal arterial bleeding using emergency DSA imaging.

Methods

Firstly, 450 patients who underwent abdominal DSA procedures were randomly selected for development of the region localization stage (RLS). Secondly, 160 consecutive patients with active abdominal arterial bleeding were included for development of the bleeding site detection stage (BSDS) and InterNet (cascade network of RLS and BSDS). Another 50 patients that ruled out active abdominal arterial bleeding were used as negative samples to evaluate InterNet performance. We evaluated the mode's efficacy using the precision-recall (PR) curve. The classification performance of a doctor with and without InterNet was evaluated using a receiver operating characteristic (ROC) curve analysis.

Results

The AP, precision, and recall of the RLS were 0.99, 0.95, and 0.99 in the validation dataset, respectively. Our InterNet reached a recall of 0.7, the precision for detection of bleeding sites was 53% in the evaluation set. The AUCs of doctors with and without InterNet were 0.803 and 0.759, respectively. In addition, the doctor with InterNet assistant could significantly reduce the elapsed time for the interpretation of each DSA sequence from 84.88 to 43.78 s.

Conclusion

Our InterNet system could assist interventional radiologists in identifying bleeding foci quickly and may improve the workflow of the DSA operation to a more real-time procedure.

Differentiation of Cerebral Dissecting Aneurysm from Hemorrhagic Saccular Aneurysm by Machine-Learning Based on Vessel Wall MRI: A Multicenter Study

The differential diagnosis of a cerebral dissecting aneurysm (DA) and a hemorrhagic saccular aneurysm (SA) often depends on the intraoperative findings; thus, improved non-invasive imaging diagnosis before surgery is essential to distinguish between these two aneurysms, in order to provide the correct formulation of surgical procedure. We aimed to build a radiomic model based on high-resolution vessel wall magnetic resonance imaging (VW-MRI) and a machine-learning algorithm. In total, 851 radiomic features from 146 cases were analyzed retrospectively, and the ElasticNet algorithm was used to establish the radiomic model in a training set of 77 cases. A clinico-radiological model using clinical features and MRI features was also built. Then an integrated model was built by combining the radiomic model and clinico-radiological model. The area under the ROC curve (AUC) was used to quantify the performance of models. The models were evaluated using leave-one-out cross-validation in a training set, and further validated in an external test set of 69 cases. The diagnostic performance of experienced radiologists was also assessed for comparison. Eight features were used to establish the radiomic model, and the radiomic model performs better (AUC = 0.831) than the clinico-radiological model (AUC = 0.717), integrated model (AUC = 0.813), and even experienced radiologists (AUC = 0.801). Therefore, a radiomic model based on VW-MRI can reliably be used to distinguish DA and hemorrhagic SA, and, thus, be widely applied in clinical practice.

Artificial Intelligence Applications in Intracranial Aneurysm: Achievements, Challenges and Opportunities

Intracranial aneurysms present in about 3% of the general population and the number of detected aneurysms is continuously rising with the advances in imaging techniques. Intracranial aneurysm rupture carries a high risk of death or permanent disabilities; therefore assessment of the intracranial aneurysm along the entire course is of great clinical importance. Given the outstanding performance of artificial intelligence (AI) in image-based tasks, many AI-based applications have emerged in recent years for the assessment of intracranial aneurysms. In this review we will summarize the state-of-the-art of AI applications in intracranial aneurysms, emphasizing the achievements, and exploring the challenges. We will also discuss the future prospects and potential opportunities. This article provides an updated view of the AI applications in intracranial aneurysms and may act as a basis for guiding the related future works.

[Development of the DSA Method for Coronary Angiography Using Deep Learning Techniques]

PURPOSE

DSA processing is not applied to the coronary artery because heavy artifacts will be generated in DSA images by heart beats and breathing of the patient. However, DSA images of coronary artery contribute to the accuracy of diagnosis and elucidating new pathogenesis of coronary artery diseases. On the other hand, deep learning techniques have been improved and shown to have a high performance. The purpose of this study was to develop a DSA method available for coronary arteries using deep learning techniques.

METHODS

We developed a convolutional neural network (CNN) -based model for this study. A total of 29 cases of coronary angiograms were used. We carried out training of the model using 21025 image patches consisting of pairs after contrast-enhanced images and before contrast-enhanced images. To obtain DSA images, we derived mask images by inputting live images to the trained model. Finally, we obtained DSA images subtracting mask images from live images. We evaluated the proposed method using subjective evaluation, which is based on human observation, and objective evaluation, which is based on standard deviation of the pixel value.

RESULT

The results of both the subjective evaluation and the objective evaluation confirmed the effectiveness of the proposed method in clinical cases.

CONCLUSION

We proposed new DSA techniques for coronary arteries using deep learning. Motion artifacts caused by heart beats and breathing were reduced, and vessel visibility was improved by the proposed DSA. By applying the technique to the area of other organs, DSA study without stopping breathe during X-ray exposure can be realized in clinical situations.

Initial investigation of predicting hematoma expansion for intracerebral hemorrhage using imaging biomarkers and machine learning

PURPOSE

Intracerebral Hemorrhage (ICH) is one of the most devastating types of strokes with mortality and morbidity rates ranging from about 51%-65% one year after diagnosis. Early hematoma expansion (HE) is a known cause of worsening neurological status of ICH patients. The goal of this study was to investigate whether non-contrast computed tomography imaging biomarkers (NCCT-IB) acquired at initial presentation can predict ICH growth in the acute stage.

MATERIALS AND METHODS

We retrospectively collected NCCT data from 200 patients with acute (<6 hours) ICH. Four NCCT-IBs (blending region, dark hole, island, and edema) were identified for each hematoma, respectively. HE status was recorded based on the clinical observation reported in the patient chart. Supervised machine learning models were developed, trained, and tested for 15 different input combinations of the NCCT-IBs to predict HE. Model performance was assessed using area under the receiver operating characteristic curve and probability for accurate diagnosis (PAD) was calculated. A 20-fold Monte-Carlo cross validation was implemented to ensure model reliability on a limited sample size of data, by running a myriad of random training/testing splits.

RESULTS

The developed algorithm was able to predict expansion utilizing all four inputs with an accuracy of 70.17%. Further testing of all biomarker combinations yielded P _AD ranging from 0.57, to 0.70.

CONCLUSION

Specific attributes of ICHs may influence the likelihood of HE and can be evaluated via a machine learning algorithm. However, certain parameters may differ in importance to reach accurate conclusions about potential expansion.

Quantitative angiography prognosis of intracranial aneurysm treatment failure using parametric imaging and distal vessel analysis

PURPOSE

Data-driven methods based on x-ray angiographic parametric imaging (API) have been successfully used to provide prognosis for intracranial aneurysm (IA) treatment outcome. Previous studies have mainly focused on embolization devices where the flow pattern visualization is in the aneurysm dome; however, this is not possible in IAs treated with endovascular coils due to high x-ray attenuation of the devices. To circumvent this challenge, we propose to investigate whether flow changes in the parent artery distal to the coil-embolized IAs could be used to achieve the same accuracy of surgical outcome prognosis.

METHODS

Eighty digital subtraction angiography sequences were acquired from patients with IA embolized with coils. Five API parameters were recorded from a region of interest (ROI) placed distal to the IA neck in the main artery. Average API values were recorded and pre-treatment values. A supervised machine learning algorithm was trained to provide a six-month post procedure binary outcome (occluded/not occluded). Receiver operating characteristic (ROC) analysis was used to assess the accuracy of the method.

RESULTS

Use of API parameters with data driven methods yielded an area under the ROC curve of 0.77 ±0.11 and accuracy of 78.6%. Single parameter-based analysis yielded accuracies which were suboptimal for clinical acceptance.

CONCLUSIONS

We determined that data-driven method based on API analysis of flow in the parent artery of IA treated with coils provide clinically acceptable accuracy for the prognosis of six months occlusion outcome.

Predicting Hematoma Expansion after Spontaneous Intracranial Hemorrhage Through a Radiomics Based Model

PURPOSE

Intracranial hemorrhage (ICH) is characterized as bleeding into the brain tissue, intracranial space, and ventricles and is the second most disabling form of stroke. Hematoma expansion (HE) following ICH has been correlated with significant neurological decline and death. For early detection of patients at risk, deep learning prediction models were developed to predict whether hematoma due to ICH will expand. This study aimed to explore the feasibility of HE prediction using a radiomic approach to help clinicians better stratify HE patients and tailor intensive therapies timely and effectively.

MATERIALS AND METHODS

Two hundred ICH patients with known hematoma evolution, were enrolled in this study. An open-source python package was utilized for the extraction of radiomic features from both non-contrast computed tomography (NCCT) and magnetic resonance imaging (MRI) scans through characterization algorithms. A total of 99 radiomic features were extracted and different features were selected for network inputs for the NCCT and MR models. Seven supervised classifiers: Support Vector Machine, Naïve Bayes, Decision Tree, Random Forest, Logistic Regression, K-Nearest Neighbor and Multilayer Perceptron were used to build the models. A training:testing split of 80:20 and 20 iterations of Monte Carlo cross validation were performed to prevent overfitting and assess the variability of the networks, respectively. The models were fed training datasets from which they learned to classify the data based on pre-determined radiomic categories.

RESULTS

The highest sensitivity among the NCCT classifier models was seen with the support vector machine (SVM) and logistic regression (LR) of 72 ± 0.3% and 73 ± 0.5%, respectively. The MRI classifier models had the highest sensitivity of 68 ± 0.5% and 72 ± 0.5% for the SVM and LR models, respectively.

CONCLUSIONS

This study indicates that the NCCT radiomics model is a better predictor of HE and that SVM and LR classifiers are better predictors of HE due to their more cautious approach indicated by a higher sensitivity metric.

Initial investigation of the use of angiographic parametric imaging for early prognosis of delayed cerebral ischemia in patients with subarachnoid hemorrhage

PURPOSE

Subarachnoid Hemorrhage (SAH) is a lethal hemorrhagic stroke that account for 25% of cerebrovascular deaths. As a result of the initial bleed, a chain of physiological events are initiated which may lead to Delayed Cerebral Ischemia (DCI). As of now we have no diagnostic capability to identify patients which may present DCI a few weeks after initial presentation. We propose to investigate whether a data driven approach using angiographic parametric imaging (API) may predict occurrence of the DCI.

MATERIALS AND METHODS

Digital Subtraction Angiographic (DSA) sequences from 125 SAH patients were used retrospectively to perform API assessment of the entire brain hemisphere where the hemorrhage was detected. Four Regions of Interests (ROIs) were placed to extract five average API biomarkers in the lateral and AP DSAs. Data driven analysis using Logistic Regression was performed for various API parameters and ROIs to find the optimal configuration to maximize the prognosis accuracy. Each model performance was evaluated using area under the curve of the receiver operator characteristic (AUROC).

RESULTS

Data driven approach with API has a 60% accuracy predicting DCI occurrence. We determined that location of the ROI for extraction of the API parameters is very important for the data driven model performance. Normalizing the values using the inlet velocities for each patient yield higher and more consistent results. Single API biomarkers models had poor prediction accuracies, barely better than chance.

CONCLUSIONS

This effectiveness exploratory study demonstrates for the first time, that prognosis of the DCI in SAH patients, is feasible and warrants a more in-depth investigation.

Recovery of complete time density curves from incomplete angiographic data using recurrent neural networks

Quantitative angiography is a 2D/3D x-ray imaging modality that summarizes hemodynamic information using time density curve (TDC) based parameters. Estimation of the TDC parameters are susceptible to errors due to various factors including, patient motion, incomplete temporal data, imaging trigger errors etc. In this study, we tested the feasibility of using recurrent neural networks (RNN) to recover complete TDC temporal information from incomplete sequences and evaluate quantitative parameters generated from the corrected TDCs. Digital subtraction angiograms (DSAs) were collected from patients undergoing endovascular treatments and angiographic parametric imaging (API) parameters were calculated from each DSA. Each set of API parameters was used to simulate a TDC resulting in a dataset of 760 TDCs. One-third of each TDC was continuously masked from pseudo-random points past the peak height (PH) point to simulate missing/artifact information. An RNN was developed, trained and tested to generate completed/corrected TDCs. The RNN recovered complete TDC temporal information with an average mean squared error of 0.0086±0.002. Average mean absolute errors were calculated between each API parameter generated from the ground truth TDCs and RNN corrected TDCs, these were 11.02%±0.91 for time to peak, 10.97%±0.69 for mean transit time, 5.65%±0.76 for PH, and 15.08%±0.98 for area under the TDC. The change in API parameters was not clinically significant and the predictive power of the API parameters was retained. This study proved the feasibility of using RNNs to mitigate motion artifacts and incomplete angiographic acquisitions to extract accurate quantitative parameters.

Detection and analysis of cerebral aneurysms based on X-ray rotational angiography - the CADA 2020 challenge

The Cerebral Aneurysm Detection and Analysis (CADA) challenge was organized to support the development and benchmarking of algorithms for detecting, analyzing, and risk assessment of cerebral aneurysms in X-ray rotational angiography (3DRA) images. 109 anonymized 3DRA datasets were provided for training, and 22 additional datasets were used to test the algorithmic solutions. Cerebral aneurysm detection was assessed using the F2 score based on recall and precision, and the fit of the delivered bounding box was assessed using the distance to the aneurysm. The segmentation quality was measured using the Jaccard index and a combination of different surface distance measures. Systematic errors were analyzed using volume correlation and bias. Rupture risk assessment was evaluated using the F2 score. 158 participants from 22 countries registered for the CADA challenge. The U-Net-based detection solutions presented by the community show similar accuracy compared to experts (F2 score 0.92), with a small number of missed aneurysms with diameters smaller than 3.5 mm. In addition, the delineation of these structures, based on U-Net variations, is excellent, with a Jaccard score of 0.92. The rupture risk estimation methods achieved an F2 score of 0.71. The performance of the detection and segmentation solutions is equivalent to that of human experts. The best results are obtained in rupture risk estimation by combining different image-based, morphological, and computational fluid dynamic parameters using machine learning methods. Furthermore, we evaluated the best methods pipeline, from detecting and delineating the vessel dilations to estimating the risk of rupture. The chain of these methods achieves an F2-score of 0.70, which is comparable to applying the risk prediction to the ground-truth delineation (0.71).

Machine Learning and Intracranial Aneurysms: From Detection to Outcome Prediction

Machine learning (ML) is a rapidly rising research tool in biomedical sciences whose applications include segmentation, classification, disease detection, and outcome prediction. With respect to traditional statistical methods, ML algorithms have the potential to learn and improve their predictive performance when fed with large data sets without the need of being specifically programmed. In recent years, this technology has been increasingly applied for tackling clinical issues in intracranial aneurysm (IA) research. Several studies attempted to provide reliable models for enhanced aneurysm detection. Convolutional neural networks trained with variable degrees of human interaction on data from diverse imaging modalities showed high sensitivity in aneurysm detection tasks, also outperforming expert image analysis. Algorithms were also shown to differentiate ruptured from unruptured IAs, with however limited clinical relevance. For prediction of rupture and stability assessment, ML was preliminarily shown to achieve better performance compared to conventional statistical methods and existing risk scores. ML-based complication and functional outcome prediction in the event of SAH have been more extensively reported, in contrast with periprocedural outcome investigation in unruptured IA patients. ML has the potential to be a game changer in IA patient management. Currently clinical translation of experimental results is limited.

Validation of an artificial intelligence-driven large vessel occlusion detection algorithm for acute ischemic stroke patients

Rapid and accurate diagnosis of large vessel occlusions (LVOs) in acute ischemic stroke (AIS) patients using automated software could improve clinical workflow in determining thrombectomy in eligible patients. Artificial intelligence-based methods could accomplish this; however, their performance in various clinical scenarios, relative to clinical experts, must be thoroughly investigated. We aimed to assess the ability of Canon's AUTOStroke Solution LVO application in properly detecting and locating LVOs in AIS patients. Data from 202 LVO and 101 non-LVO AIS patients who presented with stroke-like symptoms between March 2019 and February 2020 were collected retrospectively. LVO patients had either an internal carotid artery (ICA) (n = 59), M1 middle cerebral artery (MCA) (n = 82) or M2 MCA (n = 61) occlusion. Computed tomography angiography (CTA) scans from each patient were pushed to the automation platform and analyzed. The algorithm's ability to detect LVOs was assessed using accuracy, sensitivity and Matthews correlation coefficients (MCCs) for each occlusion type. The following results were calculated for each occlusion type in the study (accuracy, sensitivity, MCC): ICA = (0.95, 0.90, 0.89), M1 MCA = (0.89, 0.77, 0.78) and M2 MCA = (0.80, 0.51, 0.59). For the non-LVO cohort, 98% (99/101) of cases were correctly predicted as LVO negative. Processing time for each case was 69.8 ± 1.1 seconds (95% confidence interval). Canon's AUTOStroke Solution LVO application was able to accurately identify ICA and M1 MCA occlusions in addition to almost perfectly assessing when an LVO was not present. M2 MCA occlusion detection needs further improvement based on the sensitivity results displayed by the LVO detection algorithm.

Automated Collateral Flow Assessment in Patients with Acute Ischemic Stroke Using Computed Tomography with Artificial Intelligence Algorithms

BACKGROUND

Collateral circulation is associated with improved functional outcome in patients with large vessel occlusion acute ischemic stroke (AIS) who undergo reperfusion therapy. Assessment of collateral flow can be time consuming, subjective, and difficult because of complex neurovasculature. This study assessed the ability of multiple artificial intelligence algorithms in determining collateral flow of patients with AIS.

METHODS

Two hundred patients with AIS between March 2019 and January 2020 were included in this retrospective study. Peak arterial computed tomography perfusion volumes were used to assess collateral scores. Neural networks were developed for dichotomized (≥50% or <50%) and multiclass (0% filling, 0%-50% filling, 50%-100% filling, or 100% filling) collateral scoring. Maximum intensity projections from axial and anteroposterior (AP) views were synthesized for each bone subtracted three-dimensional volume and used as network inputs separately and together, along with three-dimensional data. Training:testing:validation splits of 60:30:10 and 20 iterations of Monte Carlo cross-validation were used. Network performance was assessed using 95% confidence intervals of accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).

RESULTS

The axial and AP input combination provided the most accurate results for dichotomized classification: accuracy, 0.85 ± 0.01; sensitivity, 0.88 ± 0.02; specificity, 0.82 ± 0.03; PPV, 0.86 ± 0.02; and NPV, 0.83 ± 0.03. Similarly, the axial and AP input combination provided the best results for multiclass classification: accuracy, 0.80 ± 0.01; sensitivity, 0.64 ± 0.01; specificity, 0.85 ± 0.01; PPV, 0.65 ± 0.02; and NPV, 0.85 ± 0.01.

CONCLUSIONS

This study reports one of the first artificial intelligence-based algorithms capable of accurately and efficiently assessing collateral flow of patients with AIS. This automated method for determining collateral filling could streamline clinical workflow, reduce bias, and aid in clinical decision making for determining reperfusion-eligible patients.

Using angiographic parametric imaging-derived radiomics features to predict complications and embolization outcomes of intracranial aneurysms treated by pipeline embolization devices

BACKGROUND

Pipeline embolization devices (PEDs) have gained widespread popularity in the treatment of intracranial aneurysms (IAs). However, precise predictors of treatment outcomes are still lacking. This study aimed to use angiographic parametric imaging (API)-derived radiomics features to explore whether biomarkers extracted from immediate postprocedural digital subtraction angiography (DSA) were associated with complications and embolization outcomes of IAs treated with PED without adjunctive coils.

METHODS

Radiomic features were extracted from postprocedural DSA by API, and radiomics feature selection and radiomics score calculation were performed by the least absolute shrinkage and selection operator (LASSO) logistic regression. Angiographic findings and clinical characteristics were screened using stepwise multivariable logistic regression analysis to identify significant variables for predicting the complication endpoint. Radiomics feature selection and radiomics risk score (RadRS) calculations were performed by LASSO Cox regression. Univariate and multivariate Cox regression analyses were used to identify significant predictors for the occlusion endpoint.

RESULTS

We screened 281 observations for complications and 235 observations for embolization outcomes from IAs treated in our center using PED between June 2015 and July 2020. Multivariate regression analysis showed association of the radiomics score (p<0.01) and hypertension (p=0.04) with complications. RadRS (p<0.01), symptoms (p<0.01), and age (p=0.03) were predictors of embolization outcomes. Kaplan-Meier analysis revealed that symptomatic patients (p<0.01) and those with off-label IAs (p=0.03) had shorter intervals to complete occlusion.

CONCLUSIONS

Biomarkers extracted from immediate postprocedural DSA by API could be potential indicators for assessing treatment outcomes of IAs treated by PED without adjunctive coils.

Assessment of an Artificial Intelligence Algorithm for Detection of Intracranial Hemorrhage

BACKGROUND

Immediate and accurate detection of intracranial hemorrhages (ICHs) is essential to provide a good clinical outcome for patients with ICH. Artificial intelligence has the potential to provide this, but the assessment of these methods needs to be investigated in depth. This study aimed to assess the ability of Canon's ^AUTOStroke Solution ICH detection algorithm to accurately identify patients both with and without ICHs present.

METHODS

Data from 200 ICH and 102 non-ICH patients who presented with stroke-like symptoms between August 2016 and December 2019 were collected retrospectively. Patients with ICH had at least one of the following hemorrhage types: intraparenchymal (n = 181), intraventricular (n = 45), subdural (n = 13), or subarachnoid (n = 19). Noncontrast computed tomography scans were analyzed for each patient using Canon's ^AUTOStroke Solution ICH algorithm to determine which slices contained hemorrhage. The algorithm's ability to detect ICHs was assessed using sensitivity, specificity, positive predictive value, and negative predictive value. Percentages of cases correctly identified as ICH positive and negative were additionally calculated.

RESULTS

Automated analysis demonstrated the following metrics for identifying hemorrhage slices within all 200 patients with ICH (95% confidence intervals): sensitivity = 0.93 ± 0.03, specificity = 0.93 ± 0.01, positive predictive value = 0.85 ± 0.02, and negative predictive value = 0.98 ± 0.01. A total of 95% (245 of 258) of ICH volumes were correctly triaged, whereas 88.2% (90 of 102) of non-ICH cases were correctly classified as ICH negative.

CONCLUSIONS

Canon's ^AUTOStroke Solution ICH detection algorithm was able to accurately detect intraparenchymal, intraventricular, subdural, and subarachnoid hemorrhages in addition to accurately determine when an ICH was not present. Having this automated ICH detection method could drastically improve treatment times for patients with ICH.

The Aneurysm Occlusion Assistant, an AI platform for real time surgical guidance of intracranial aneurysms

Purpose

In recent years, endovascular treatment has become the dominant approach to treat intracranial aneurysms (IAs). Despite tremendous improvement in surgical devices and techniques, 10-30% of these surgeries require retreatment. Previously, we developed a method which combines quantitative angiography with data-driven modeling to predict aneurysm occlusion within a fraction of a second. This is the first report on a semi-autonomous system, which can predict the surgical outcome of an IA immediately following device placement, allowing for therapy adjustment. Additionally, we previously reported various algorithms which can segment IAs, extract hemodynamic parameters via angiographic parametric imaging, and perform occlusion predictions.

Methods

We integrated these features into an Aneurysm Occlusion Assistant (AnOA) utilizing the Kivy library's graphical instructions and unique language properties for interface development, while the machine learning algorithms were entirely developed within Keras, Tensorflow and skLearn. The interface requires pre- and post-device placement angiographic data. The next steps for aneurysm segmentation, angiographic analysis and prediction have been integrated allowing either autonomous or interactive use.

Results

The interface allows for segmentation of IAs and cranial vasculature with a dice index of ~0.78 and prediction of aneurysm occlusion at six months with an accuracy 0.84, in 6.88 seconds.

Conclusion

This is the first report on the AnOA to guide endovascular treatment of IAs. While this initial report is on a stand-alone platform, the software can be integrated in the angiographic suite allowing direct communication with the angiographic system for a completely autonomous surgical guidance solution.

Use of biplane quantitative angiographic imaging with ensemble neural networks to assess reperfusion status during mechanical thrombectomy

Digital subtraction angiography (DSA) is the main imaging modality used to assess reperfusion during mechanical thrombectomy (MT) when treating large vessel occlusion (LVO) ischemic strokes. To improve this visual and subjective assessment, hybrid models combining angiographic parametric imaging (API) with deep learning tools have been proposed. These models use convolutional neural networks (CNN) with single view individual API maps, thus restricting use of complementary information from multiple views and maps resulting in loss of relevant clinical information. This study investigates use of ensemble networks to combine hemodynamic information from multiple bi-plane API maps to assess level of reperfusion. Three-hundred-eighty-three anteroposterior (AP) and lateral view DSAs were retrospectively collected from patients who underwent MTs of anterior circulation LVOs. API peak height (PH) and area under time density curve (AUC) maps were generated. CNNs were developed to classify maps as adequate/inadequate reperfusion as labeled by two neuro-interventionalists. Outputs from individual networks were combined by weighting each output, using a grid search algorithm. Ensembled, AP-AUC, AP-PH, lateral-AUC, and lateral-PH networks achieved accuracies of 83.0% (95% confidence-interval: 81.2%-84.8%), 74.4% (72.0%-76.7%), 74.2% (72.8%-75.7%), 74.9% (72.2%-77.7%), and 76.9% (74.4%-79.5%); area under receiver operating characteristic curves of 0.86 (0.84-0.88), 0.81 (0.79-0.83), 0.83 (0.81-0.84), 0.82 (0.8-0.84), and 0.84 (0.82-0.87); and Matthews correlation coefficients of 0.66 (0.63-0.70), 0.48 (0.43-0.53), 0.49 (0.46-0.52), 0.51 (0.45-0.56), and 0.54 (0.49-0.59) respectively. Ensembled network performance was significantly better than individual networks (McNemar's p-value<0.05). This study proved feasibility of using ensemble networks to combine hemodynamic information from multiple bi-plane API maps to assess level of reperfusion during MTs.

Investigation of the efficacy of a data-driven CT artifact correction scheme for sparse and truncated projection data for intracranial hemorrhage diagnosis

Data-driven CT-image reconstruction techniques for truncated or sparsely acquired projection data to reduce radiation dose, iodine volume, and patient motion artifacts have been proposed. These approaches have shown good performance and preservation of image quality metrics. To continue these efforts, we investigated whether these techniques affect the performance of a machine-learning algorithm to identify the presence of intracranial hemorrhage (ICH). Ten-thousand head CT scans were collected from the 2019 RSNA Intracranial Hemorrhage Detection and Classification Challenge dataset. Sinograms were simulated and then resampled in both a one-third truncated and one-third sparse manner. GANs were tasked with correcting the incomplete projection data in two ways. Firstly, in the sinogram domain, where the incomplete sinogram was filled by the GAN and then reconstructed. Secondly, in the reconstruction domain, where the incomplete data were first reconstructed and the sparse or truncation artifacts were corrected by the GAN. Eighty-five hundred images were used for artifact correction network training, and 1500 were withheld for network assessment via an already trained machine-learning algorithm tasked with diagnosis of ICH presence. Fully-sampled reconstructions were compared with the sparse and truncated reconstructions for classification accuracy. Difference in classification accuracy between the fully sampled (83.4%), sparse (82.0%), and truncated (82.3%) reconstructions was minimal, demonstrating that the network diagnosis performance is unaffected by 2/3 reduction of projection data. This work indicates that data-driven reconstructions for a sparse or truncated projection dataset can provide high diagnostic performance for ICH detection at a fraction of the typical radiation dose.

Application of radiomics and machine learning in head and neck cancers

With the continuous development of medical image informatics technology, more and more high-throughput quantitative data could be extracted from digital medical images, which has resulted in a new kind of omics-Radiomics. In recent years, in addition to genomics, proteomics and metabolomics, radiomic has attracted the interest of more and more researchers. Compared to other omics, radiomics can be perfectly integrated with clinical data, even with the pathology and molecular biomarker, so that the study can be closer to the clinical reality and more revealing of the tumor development. Mass data will also be generated in this process. Machine learning, due to its own characteristics, has a unique advantage in processing massive radiomic data. By analyzing mass amounts of data with strong clinical relevance, people can construct models that more accurately reflect tumor development and progression, thereby providing the possibility of personalized and sequential treatment of patients. As one of the cancer types whose treatment and diagnosis rely on imaging examination, radiomics has a very broad application prospect in head and neck cancers (HNC). Until now, there have been some notable results in HNC. In this review, we will introduce the concepts and workflow of radiomics and machine learning and their current applications in head and neck cancers, as well as the directions and applications of artificial intelligence in the treatment and diagnosis of HNC.

CT artifact correction for sparse and truncated projection data using generative adversarial networks

PURPOSE

Computed tomography image reconstruction using truncated or sparsely acquired projection data to reduce radiation dose, iodine volume, and patient motion artifacts has been widely investigated. To continue these efforts, we investigated the use of machine learning-based reconstruction techniques using deep convolutional generative adversarial networks (DCGANs) and evaluated its effect using standard imaging metrics.

METHODS

Ten thousand head computed tomography (CT) scans were collected from the 2019 RSNA Intracranial Hemorrhage Detection and Classification Challenge dataset. Sinograms were simulated and then resampled in both a one-third truncated and one-third sparse manner. DCGANs were tasked with correcting the incomplete projection data, either in the sinogram domain where the full sinogram was recovered by the DCGAN and then reconstructed, or the reconstruction domain where the incomplete data were first reconstructed and the sparse or truncation artifacts were corrected by the DCGAN. Seventy-five hundred images were used for network training and 2500 were withheld for network assessment using mean absolute error (MAE), structural similarity index measure (SSIM), and peak signal-to-noise ratio (PSNR) between results of different correction techniques. Image data from a quality-assurance phantom were also resampled in the two manners and corrected and reconstructed for network performance assessment using line profiles across high-contrast features, the modulation transfer function (MTF), noise power spectrum (NPS), and Hounsfield Unit (HU) linearity analysis.

RESULTS

Better agreement with the fully sampled reconstructions were achieved from sparse acquisition corrected in the sinogram domain and the truncated acquisition corrected in the reconstruction domain. MAE, SSIM, and PSNR showed quantitative improvement from the DCGAN correction techniques. HU linearity of the reconstructions was maintained by the correction techniques for the sparse and truncated acquisitions. MTF curves reached the 10% modulation cutoff frequency at 5.86 lp/cm for the truncated corrected reconstruction compared with 2.98 lp/cm for the truncated uncorrected reconstruction, and 5.36 lp/cm for the sparse corrected reconstruction compared with around 2.91 lp/cm for the sparse uncorrected reconstruction. NPS analyses yielded better agreement across a range of frequencies between the resampled corrected phantom and truth reconstructions.

CONCLUSIONS

We demonstrated the use of DCGANs for CT-image correction from sparse and truncated simulated projection data, while preserving imaging quality of the fully sampled projection data.

Predictive score for complete occlusion of intracranial aneurysms treated by flow-diverter stents using machine learning

BACKGROUND

Complete occlusion of an intracranial aneurysm (IA) after the deployment of a flow-diverter stent is currently unpredictable. The aim of this study was to develop a predictive occlusion score based on pretreatment clinical and angiographic criteria.

METHODS

Consecutive patients with ≥6 months follow-up were included from 2008 to 2019 and retrospectively analyzed. Each IA was evaluated using the Raymond-Roy occlusion classification (RROC) and dichotomized as occluded (A) or residual (B/C); 80% of patients were randomly assigned to the training sample. Feature selection and binary outcome prediction relied on logistic regression and threshold maximizing class separation selected by a CART tree algorithm. The feature selection was addressed by a genetic algorithm selected from the 30 pretreatment available variables.

RESULTS

The study included 146 patients with 154 IAs. Feature selection yielded a combination of six variables with a good cross-validated accuracy on the test sample, a combination we labeled DIANES score (IA diameter, indication, parent artery diameter ratio, neck ratio, side-branch artery, and sex). A score of more than -6 maximized the ability to predict RROC=A with sensitivity of 87% (95% CI 79% to 95%) and specificity of 82% (95% CI 64% to 96%) in the training sample. Accuracy was 86% (95% CI 79% to 94%). In the test sample, sensitivity and specificity were 89% (95% CI 77% to 98%) and 60% (95% CI 33% to 86%), respectively. Accuracy was 81% (95% CI 69% to 91%).

CONCLUSION

A score was developed as a grading scale for prediction of the final occlusion status of IAs treated with a flow-diverter stent.

Performance of Radiomics derived morphological features for prediction of aneurysm rupture status

BACKGROUND

Morphological differences between ruptured and unruptured cerebral aneurysms represent a focus of neuroimaging researchfor understanding the mechanisms of aneurysmal rupture. We evaluated the performance of Radiomics derived morphological features, recently proposed for rupture status classification, against automatically measured shape and size features previously established in the literature.

METHODS

353 aneurysms (123 ruptured) from three-dimensional rotational catheter angiography (3DRA) datasets were analyzed. Based on a literature review, 13 Radiomics and 13 established morphological descriptors were automatically extracted per aneurysm, and evaluated for rupture status prediction using univariate and multivariate statistical analysis, yielding an area under the curve (AUC) metric of the receiver operating characteristic.

RESULTS

Validation of overlapping descriptors for size/volume using both methods were highly correlated (p<0.0001, R ²=0.99). Univariate analysis selected AspectRatio (p<0.0001, AUC=0.75), Non-sphericity Index (p<0.0001, AUC=0.75), Height/Width (p<0.0001, AUC=0.73), and SizeRatio (p<0.0001, AUC=0.73) as best among established descriptors, and Elongation (p<0.0001, AUC=0.71) and Flatness (p<0.0001, AUC=0.72) among Radiomics features. Radiomics Elongation correlated best with established Height/Width (R ²=0.52), whereas Radiomics Flatness correlated best with Ellipticity Index (R ²=0.54). Radiomics Sphericity correlated best with Undulation Index (R ²=0.65). Best Radiomics performers, Elongation and Flatness, were highly correlated descriptors (p<0.0001, R ²=0.75). In multivariate analysis, established descriptors (Height/Width, SizeRatio, Ellipticity Index; AUC=0.79) outperformed Radiomics features (Elongation, Maximum3Ddiameter; AUC=0.75).

CONCLUSION

Although recently introduced Radiomics analysis for aneurysm shape and size evaluation has the advantage of being an efficient operator independent methodology, it currently offers inferior rupture status discriminant performance compared with established descriptors. Future research is needed to extend the current Radiomics feature set to better capture aneurysm shape information.

Deep learning based detection of intracranial aneurysms on digital subtraction angiography: A feasibility study

BACKGROUND

Digital subtraction angiography is the gold standard for detecting and characterising aneurysms. Here, we assess the feasibility of commercial-grade deep learning software for the detection of intracranial aneurysms on whole-brain anteroposterior and lateral 2D digital subtraction angiography images.

MATERIAL AND METHODS

Seven hundred and six digital subtraction angiography images were included from a cohort of 240 patients (157 female, mean age 59 years, range 20-92; 83 male, mean age 55 years, range 19-83). Three hundred and thirty-five (47%) single frame anteroposterior and lateral images of a digital subtraction angiography series of 187 aneurysms (41 ruptured, 146 unruptured; average size 7±5.3 mm, range 1-5 mm; total 372 depicted aneurysms) and 371 (53%) aneurysm-negative study images were retrospectively analysed regarding the presence of intracranial aneurysms. The 2D data was split into testing and training sets in a ratio of 4:1 with 3D rotational digital subtraction angiography as gold standard. Supervised deep learning was performed using commercial-grade machine learning software (Cognex, ViDi Suite 2.0). Monte Carlo cross validation was performed.

RESULTS

Intracranial aneurysms were detected with a sensitivity of 79%, a specificity of 79%, a precision of 0.75, a F1 score of 0.77, and a mean area-under-the-curve of 0.76 (range 0.68-0.86) after Monte Carlo cross-validation, run 45 times.

CONCLUSION

The commercial-grade deep learning software allows for detection of intracranial aneurysms on whole-brain, 2D anteroposterior and lateral digital subtraction angiography images, with results being comparable to more specifically engineered deep learning techniques.

Feasibility study for use of angiographic parametric imaging and deep neural networks for intracranial aneurysm occlusion prediction

Background

Angiographic parametric imaging (API), based on digital subtraction angiography (DSA), is a quantitative imaging tool that may be used to extract contrast flow parameters related to hemodynamic conditions in abnormal pathologies such as intracranial aneurysms (IAs).

Objective

To investigate the feasibility of using deep neural networks (DNNs) and API to predict IA occlusion using pre- and post-intervention DSAs.

Methods

We analyzed DSA images of IAs pre- and post-treatment to extract API parameters in the IA dome and the corresponding main artery (un-normalized data). We implemented a two-step correction to account for injection variability (normalized data) and projection foreshortening (relative data). A DNN was trained to predict a binary IA occlusion outcome: occluded/unoccluded. Network performance was assessed with area under the receiver operating characteristic curve (AUROC) and classification accuracy. To evaluate the effect of the proposed corrections, prediction accuracy analysis was performed after each normalization step.

Results

The study included 190 IAs. The mean and median duration between treatment and follow-up was 9.8 and 8.0 months, respectively. For the un-normalized, normalized, and relative subgroups, the DNN average prediction accuracies for IA occlusion were 62.5% (95% CI 60.5% to 64.4%), 70.8% (95% CI 68.2% to 73.4%), and 77.9% (95% CI 76.2% to 79.6%). The average AUROCs for the same subgroups were 0.48 (0.44–0.52), 0.67 (0.61–0.73), and 0.77 (0.74–0.80).

Conclusions

The study demonstrated the feasibility of using API and DNNs to predict IA occlusion using only pre- and post-intervention angiographic information.