Detection Of Acute Intracranial Hemorrhage Research Articles

Predicting Patient Demographics From Chest Radiographs With Deep Learning.

Deep learning models are increasingly informing medical decision making, for instance, in the detection of acute intracranial hemorrhage and pulmonary embolism. However, many models are trained on medical image databases that poorly represent the diversity of the patients they serve. In turn, many artificial intelligence models may not perform as well on assisting providers with important medical decisions for underrepresented populations. Assessment of the ability of deep learning models to classify the self-reported gender, age, self-reported ethnicity, and insurance status of an individual patient from a given chest radiograph. Models were trained and tested with 55,174 radiographs in the MIMIC Chest X-ray (MIMIC-CXR) database. External validation data came from two separate databases, one from CheXpert and another from a multihospital urban health care system after institutional review board approval. Macro-averaged area under the curve (AUC) values were used to evaluate performance of models. Code used for this study is open-source and available at https://github.com/ai-bias/cxr-bias, and pixelstopatients.com/models/demographics. Accuracy of models to predict gender was nearly perfect, with 0.999 (95% confidence interval: 0.99-0.99) AUC on held-out test data and 0.994 (0.99-0.99) and 0.997 (0.99-0.99) on external validation data. There was high accuracy to predict age and ethnicity, ranging from 0.854 (0.80-0.91) to 0.911 (0.88-0.94) AUC, and moderate accuracy to predict insurance status, with AUC ranging from 0.705 (0.60-0.81) on held-out test data to 0.675 (0.54-0.79) on external validation data. Deep learning models can predict the age, self-reported gender, self-reported ethnicity, and insurance status of a patient from a chest radiograph. Visualization techniques are useful to ensure deep learning models function as intended and to demonstrate anatomical regions of interest. These models can be used to ensure that training data are diverse, thereby ensuring artificial intelligence models that work on diverse populations.

Abstract 20: Deep Learning-Based Automated Intracranial Hemorrhage Detection and Notification

Introduction: Ruling out hemorrhage (stroke or traumatic) prior to administration of thrombolytics is critical for Code Strokes. A triage software that identifies hemorrhages on head CTs and alerts radiologists would help to streamline patient care and increase diagnostic confidence and patient safety. ML approach: We trained a deep convolutional network with a hybrid 3D/2D architecture on unenhanced head CTs of 805 patients. Our training dataset comprised 348 positive hemorrhage cases (IPH=245, SAH=67, Sub/Epi-dural=70, IVH=83) (128 female) and 457 normal controls (217 female). Lesion outlines were drawn by experts and stored as binary masks that were used as ground truth data during the training phase (random 80/20 train/test split). Diagnostic sensitivity and specificity were defined on a per patient study level, i.e. a single, binary decision for presence/absence of a hemorrhage on a patient’s CT scan. Final validation was performed in 380 patients (167 positive). Tool: The hemorrhage detection module was prototyped in Python/Keras. It runs on a local LINUX server (4 CPUs, no GPUs) and is embedded in a larger image processing platform dedicated to stroke. Results: Processing time for a standard whole brain CT study (3-5mm slices) was around 2min. Upon completion, an instant notification (by email and/or mobile app) was sent to users to alert them about the suspected presence of a hemorrhage. Relative to neuroradiologist gold standard reads the algorithm’s sensitivity and specificity is 90.4% and 92.5% (95% CI: 85%-94% for both). Detection of acute intracranial hemorrhage can be automatized by deploying deep learning. It yielded very high sensitivity/specificity when compared to gold standard reads by a neuroradiologist. Volumes as small as 0.5mL could be detected reliably in the test dataset. The software can be deployed in busy practices to prioritize worklists and alert health care professionals to speed up therapeutic decision processes and interventions.

Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning

Computed tomography (CT) of the head is used worldwide to diagnose neurologic emergencies. However, expertise is required to interpret these scans, and even highly trained experts may miss subtle life-threatening findings. For head CT, a unique challenge is to identify, with perfect or near-perfect sensitivity and very high specificity, often small subtle abnormalities on a multislice cross-sectional (three-dimensional [3D]) imaging modality that is characterized by poor soft tissue contrast, low signal-to-noise using current low radiation-dose protocols, and a high incidence of artifacts. We trained a fully convolutional neural network with 4,396 head CT scans performed at the University of California at San Francisco and affiliated hospitals and compared the algorithm's performance to that of 4 American Board of Radiology (ABR) certified radiologists on an independent test set of 200 randomly selected head CT scans. Our algorithm demonstrated the highest accuracy to date for this clinical application, with a receiver operating characteristic (ROC) area under the curve (AUC) of 0.991 ± 0.006 for identification of examinations positive for acute intracranial hemorrhage, and also exceeded the performance of 2 of 4 radiologists. We demonstrate an end-to-end network that performs joint classification and segmentation with examination-level classification comparable to experts, in addition to robust localization of abnormalities, including some that are missed by radiologists, both of which are critically important elements for this application.

Advances of MRI in stroke

The imaging of stroke has several potential objectives: 1. Differentiation of stroke from non-stroke. 2. Classification of stroke sub-types. 3. Identification of patients likely to benefit from acute interventions. 4. Identification of patients at risk from stroke treatment. 5. Guidance of in-patient management. 6. Selection of appropriate secondary prevention strategies. 7. Research to improve understanding of stroke pathology, the effect in interventions, and the development and testing of new treatments. For most indications, MRI offers advantages over CT scanning. The two modalities are equivalently sensitive to detection of acute intracranial hemorrhage, but MRI is more sensitive to and more precisely delineates a variety of brain pathologies. In particular, MRI is superior for the detection of the earlier and smaller regions of ischemic changes using diffusion weighted imaging (DWI), and for the presence of chronic microhemorrhages, which are indicative of future risk of intracranial hemorrhage. The most important recent therapeutic advances in imaging-based acute stroke intervention have expanded the time window for treatable ischemic stroke. Clinical trials proved the benefit of intravenous alteplase for ischemic stroke in patients who wake-up with stroke symptoms or otherwise present later than the standard time from last known well, if MRI shows a positive DWI but negative FLAIR, indicative of early stroke, or shows a diffusion-perfusion mismatch indicative of ischemic penumbra. A penumbral pattern on MRI in the presence of acute occlusion of the internal carotid or middle cerebral artery identifies patients who benefit from mechanical thrombectomy 6-24 hours from time last known well.

Best Paper Selection

Bandeira Diniz JO, Bandeira Diniz PH, Azevedo Valente TL, Corrêa Silva A, de Paiva AC, Gattass M. Detection of mass regions in mammograms by bilateral analysis adapted to breast density using similarity indexes and convolutional neural networks. Comput Methods Programs Biomed 2018 Mar;156:191-207 https://www.sciencedirect.com/science/article/pii/S0169260717304248?via%3Dihub Lee H, Yune S, Mansouri M, Ki M, Tajmir SH, Guerrier CE, Ebert SA, Pomerantz SR, Romero JM, Kamalian S, Gonzalez RG, Lev MH, Do S. An explainable deep-learning algorithm for the detection of acute intracranial hemorrhage from small datasets. Nat Biomed Eng 2019 Mar;3(3):173-82 https://www.nature.com/articles/s41551-018-0324-9 Samad MD, Ulloa A, Wehner GJ, Jing L, Hartzel D, Good CW, Williams BA, Haggerty CM, Fornwalt BK. Predicting survival from large echocardiography and electronic health record datasets: optimization with machine learning. JACC Cardiovasc Imaging 2019 Apr;12(4):681-9 https://www.sciencedirect.com/science/article/pii/S1936878X18303851?via%3Dihub Vasilakakis MD, Iakovidis DK, Spyrou E, Koulaouzidis A. DINOSARC: color features based on selective aggregation of chromatic image components for wireless capsule endoscopy. Comput Math Methods Med 2018 Sep 3;2018:2026962 https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6140007/

Multi-resolution statistical image reconstruction for mitigation of truncation effects: application to cone-beam CT of the head

A prototype cone-beam CT (CBCT) head scanner featuring model-based iterative reconstruction (MBIR) has been recently developed and demonstrated the potential for reliable detection of acute intracranial hemorrhage (ICH), which is vital to diagnosis of traumatic brain injury and hemorrhagic stroke. However, data truncation (e.g. due to the head holder) can result in artifacts that reduce image uniformity and challenge ICH detection. We propose a multi-resolution MBIR method with an extended reconstruction field of view (RFOV) to mitigate truncation effects in CBCT of the head. The image volume includes a fine voxel size in the (inner) nontruncated region and a coarse voxel size in the (outer) truncated region. This multi-resolution scheme allows extension of the RFOV to mitigate truncation effects while introducing minimal increase in computational complexity. The multi-resolution method was incorporated in a penalized weighted least-squares (PWLS) reconstruction framework previously developed for CBCT of the head. Experiments involving an anthropomorphic head phantom with truncation due to a carbon-fiber holder were shown to result in severe artifacts in conventional single-resolution PWLS, whereas extending the RFOV within the multi-resolution framework strongly reduced truncation artifacts. For the same extended RFOV, the multi-resolution approach reduced computation time compared to the single-resolution approach (viz. time reduced by 40.7%, 83.0%, and over 95% for an image volume of 6003, 8003, 10003 voxels). Algorithm parameters (e.g. regularization strength, the ratio of the fine and coarse voxel size, and RFOV size) were investigated to guide reliable parameter selection. The findings provide a promising method for truncation artifact reduction in CBCT and may be useful for other MBIR methods and applications for which truncation is a challenge.

Modeling and design of a cone-beam CT head scanner using task-based imaging performance optimization

Detection of acute intracranial hemorrhage (ICH) is important for diagnosis and treatment of traumatic brain injury, stroke, postoperative bleeding, and other head and neck injuries. This paper details the design and development of a cone-beam CT (CBCT) system developed specifically for the detection of low-contrast ICH in a form suitable for application at the point of care. Recognizing such a low-contrast imaging task to be a major challenge in CBCT, the system design began with a rigorous analysis of task-based detectability including critical aspects of system geometry, hardware configuration, and artifact correction.The imaging performance model described the three-dimensional (3D) noise-equivalent quanta using a cascaded systems model that included the effects of scatter, scatter correction, hardware considerations of complementary metal-oxide semiconductor (CMOS) and flat-panel detectors (FPDs), and digitization bit depth. The performance was analyzed with respect to a low-contrast (40–80 HU), medium-frequency task representing acute ICH detection. The task-based detectability index was computed using a non-prewhitening observer model. The optimization was performed with respect to four major design considerations: (1) system geometry (including source-to-detector distance (SDD) and source-to-axis distance (SAD)); (2) factors related to the x-ray source (including focal spot size, kVp, dose, and tube power); (3) scatter correction and selection of an antiscatter grid; and (4) x-ray detector configuration (including pixel size, additive electronics noise, field of view (FOV), and frame rate, including both CMOS and a-Si:H FPDs). Optimal design choices were also considered with respect to practical constraints and available hardware components.The model was verified in comparison to measurements on a CBCT imaging bench as a function of the numerous design parameters mentioned above. An extended geometry (SAD = 750 mm, SDD = 1100 mm) was found to be advantageous in terms of patient dose (20 mGy) and scatter reduction, while a more isocentric configuration (SAD = 550 mm, SDD = 1000 mm) was found to give a more compact and mechanically favorable configuration with minor tradeoff in detectability. An x-ray source with a 0.6 mm focal spot size provided the best compromise between spatial resolution requirements and x-ray tube power. Use of a modest anti-scatter grid (8:1 GR) at a 20 mGy dose provided slight improvement (~5–10%) in the detectability index, but the benefit was lost at reduced dose. The potential advantages of CMOS detectors over FPDs were quantified, showing that both detectors provided sufficient spatial resolution for ICH detection, while the former provided a potentially superior low-dose performance, and the latter provided the requisite FOV for volumetric imaging in a centered-detector geometry.Task-based imaging performance modeling provides an important starting point for CBCT system design, especially for the challenging task of ICH detection, which is somewhat beyond the capabilities of existing CBCT platforms. The model identifies important tradeoffs in system geometry and hardware configuration, and it supports the development of a dedicated CBCT system for point-of-care application. A prototype suitable for clinical studies is in development based on this analysis.

WE‐EF‐207‐03: Design and Optimization of a CBCT Head Scanner for Detection of Acute Intracranial Hemorrhage

Purpose:To design a dedicated x‐ray cone‐beam CT (CBCT) system suitable to deployment at the point‐of‐care and offering reliable detection of acute intracranial hemorrhage (ICH), traumatic brain injury (TBI), stroke, and other head and neck injuries.Methods:A comprehensive task‐based image quality model was developed to guide system design and optimization of a prototype head scanner suitable to imaging of acute TBI and ICH. Previously reported models were expanded to include the effects of x‐ray scatter correction necessary for detection of low contrast ICH and the contribution of bit depth (digitization noise) to imaging performance. Task‐based detectablity index provided the objective function for optimization of system geometry, x‐ray source, detector type, anti‐scatter grid, and technique at 10–25 mGy dose. Optimal characteristics were experimentally validated using a custom head phantom with 50 HU contrast ICH inserts imaged on a CBCT imaging bench allowing variation of system geometry, focal spot size, detector, grid selection, and x‐ray technique.Results:The model guided selection of system geometry with a nominal source‐detector distance 1100 mm and optimal magnification of 1.50. Focal spot size ∼0.6 mm was sufficient for spatial resolution requirements in ICH detection. Imaging at 90 kVp yielded the best tradeoff between noise and contrast. The model provided quantitation of tradeoffs between flat‐panel and CMOS detectors with respect to electronic noise, field of view, and readout speed required for imaging of ICH. An anti‐scatter grid was shown to provide modest benefit in conjunction with post‐acquisition scatter correction. Images of the head phantom demonstrate visualization of millimeter‐scale simulated ICH.Conclusions:Performance consistent with acute TBI and ICH detection is feasible with model‐based system design and robust artifact correction in a dedicated head CBCT system. Further improvements can be achieved with incorporation of model‐based iterative reconstruction techniques also within the scope of the task‐based optimization framework.David Foos and Xiaohui Wang are employees of Carestream Health

A Lancet Issue Devoted to Stroke

Marc Fisher MD Kennedy Lees MD Section Editors: The dominant theme of the January 27, 2007 issue of Lancet is stroke. In a new editorial policy the Lancet decided to publish themed issues of particular interest to both general clinicians and specialist researchers. Stroke is undoubtedly a disease that requires special attention: approximately 1% of the 6.5 billion people living on our planet—twice the population of Canada or 8 times the population of Switzerland—will die each year of stroke. Many of these strokes are either preventable or could be delayed to a more advanced age. Considering such figures, the Lancet editors have to be congratulated for their decision, which is an important step in the struggle to reduce the burden of stroke in our societies. They have carefully selected original articles, seminars, and reviews on stroke. In the following we will comment on the 4 original articles. ### Wahlgren N, Ahmed N, Davalos A, Ford GA, Grond M, Hacke W, Hennerici MG, Kaste M, Kuelkens S, Larrue V, Lees KR, Roine RO, Soinne L, Toni D, Vanhooren G, for the SITS-MOST investigators. Thrombolysis with alteplase for acute ischemic stroke in the Safe Implementation of Thrombolysis in Stroke-Monitoring Study (SITS-MOST): an observational study. Lancet. 2007;369:275–282. The European Agency for the Evaluation of Medicinal Products (EMEA) mandated a postlicensing observational registry of thrombolysis use in Europe as a condition of licensing. The web-based SITS-MOST registry was launched in response to this need and collected more than 6000 patients from nearly 300 European centers. #### Methods Any European center was eligible as long as they agreed to enter all consecutive patients and agreed to site monitoring and aggregrate data reporting. Both experienced and inexperienced thrombolysis sites were eligible. The primary outcomes were safety outcomes: symptomatic ICH and death at 3 months from treatment. #### Results Patients enrolled in the registry were typically similar to those enrolled in prior trials and stroke thrombolysis registries. The mortality rate was 11.3% and symptomatic ICH rate 7.3%. Excellent functional outcome was achieved in 39% of patients. #### Interpretation This experience represents the largest stroke thrombolysis registry globally. It documents the safety and effectiveness of alteplase for stroke in routine clinical …

Computer aided detection of small acute intracranial hemorrhage on computer tomography of brain

Introduction Detection of acute intracranial hemorrhage (AIH) is a primary task in image interpretation of computer tomography (CT) of brain for patients suffering from acute neurological disturbance or head injury. Although CT readily depicts AIH, interpretation can be difficult especially when the lesion is inconspicuous or the reader is inexperienced. Objective To develop a computer aided detection system that improves diagnostic accuracy of small AIH on brain CT. Materials and methods Intracranial contents are first segmented by thresholding and morphological operations, which are then subjected to denoising and adjustment for CT cupping artifacts. The brain is then automatically realigned into normal position. AIH candidates are extracted based on top-hat transformation and left–right asymmetry. AIH candidates are registered against a normalized coordinate system such that the candidates are rendered anatomical information. True AIH is differentiated from mimicking normal variants or artifacts by a knowledge-based classification system incorporating rules that make use of quantified imaging features and anatomical information. A total of 186 clinical cases, including 62 CT studies showing small (<1 cm) AIH, and 124 controls, were retrospectively collected. Forty positive cases and 80 controls were used for the training of the CAD. Twenty-two positive cases and 44 controls were used in the validation of the CAD system. Regions of AIH identified by two experienced radiologists were used as gold standard. The size of individual AIH volume was also recorded. Results On a per patient basis, the system achieved sensitivity of 95% (38/40) and specificity of 88.8% (71/80) in the training dataset. The sensitivity and specificity were 100% (22/22) and 84.1% (37/44) respectively for the diagnosis of AIH in the validation cases. Individual cases contained variable number of AIH volumes. There were 77 lesions in the 40 training cases and 46 lesions in the 22 validation cases. On a per lesion basis, the sensitivities were 84.4% (65/77) and 82.6% (38/46) for all lesions 10 mm or smaller for the training and validation datasets, respectively. False positive rates were 0.19 (23/120) and 0.29 (19/66) false positive lesion per case for the training and validation datasets, respectively. Conclusion This study demonstrated that CAD is valuable for detection of small AIH on brain CT.

Comparison of Five MR Sequences for the Detection of Acute Intracranial Hemorrhage

Purpose: To evaluate the respective value of 5 MR sequences to distinguish between stroke patients with and without acute intracerebral hemorrhage (ICH). Methods: MR images obtained ≤6 h after stroke onset of 86 patients (43 ICH, 43 non-ICH) were reviewed by 3 observers who looked for signs of acute ICH on each image set [T₁ and T₂ gradient echo (GRE), FLAIR, T₂-EPI and DWI], presented separately. Results: For the identification of ICH, intraobserver and interobserver concordance were at least ĸ = 0.95 for all sequences. Of all interpretations, 7 (0.4%) were erroneous, with sensitivity and specificity of FLAIR, T₂-EPI and DWI reaching 100%; GRE sensitivity and specificity were 100% and 95–97.5%, respectively; T₁ sensitivity and specificity were 97.3–100% and 97.4–100%, respectively. All 4 patients misclassified on one pulse sequence were correctly classified on all the other sequences. Conclusion: In the setting of acute stroke, each of the 5 studied sequences enables ICH and non-ICH patients to be distinguished with high sensitivity and specificity.