A real-world demonstration of machine learning generalizability in the detection of intracranial hemorrhage on head computerized tomography.
Ontology highlight
ABSTRACT:
SUBMITTER: Salehinejad H
PROVIDER: S-EPMC8382750 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:Intracranial hemorrhage (ICH) requires prompt diagnosis to optimize patient outcomes. We hypothesized that machine learning algorithms could automatically analyze computed tomography (CT) of the head, prioritize radiology worklists and reduce time to diagnosis of ICH. 46,583 head CTs (~2 million images) acquired from 2007-2017 were collected from several facilities across Geisinger. A deep convolutional neural network was trained on 37,074 studies and subsequently evaluated on 9499 unseen studies. The predictive model was implemented prospectively for 3 months to re-prioritize "routine" head CT studies as "stat" on realtime radiology worklists if an ICH was detected. Time to diagnosis was compared between the re-prioritized "stat" and "routine" studies. A neuroradiologist blinded to the study reviewed false positive studies to determine whether the dictating radiologist overlooked ICH. The model achieved an area under the ROC curve of 0.846 (0.837-0.856). During implementation, 94 of 347 "routine" studies were re-prioritized to "stat", and 60/94 had ICH identified by the radiologist. Five new cases of ICH were identified, and median time to diagnosis was significantly reduced (p < 0.0001) from 512 to 19 min. In particular, one outpatient with vague symptoms on anti-coagulation was found to have an ICH which was treated promptly with reversal of anticoagulation, resulting in a good clinical outcome. Of the 34 false positives, the blinded over-reader identified four probable ICH cases overlooked in original interpretation. In conclusion, an artificial intelligence algorithm can prioritize radiology worklists to reduce time to diagnosis of new outpatient ICH by 96% and may also identify subtle ICH overlooked by radiologists. This demonstrates the positive impact of advanced machine learning in radiology workflow optimization.
| S-EPMC6550144 | biostudies-literature
Project description:A neural network model was previously developed to predict melatonin rhythms accurately from blue light and skin temperature recordings in individuals on a fixed sleep schedule. This study aimed to test the generalizability of the model to other sleep schedules, including rotating shift work. Ambulatory wrist blue light irradiance and skin temperature data were collected in 16 healthy individuals on fixed and habitual sleep schedules, and 28 rotating shift workers. Artificial neural network models were trained to predict the circadian rhythm of (i) salivary melatonin on a fixed sleep schedule; (ii) urinary aMT6s on both fixed and habitual sleep schedules, including shift workers on a diurnal schedule; and (iii) urinary aMT6s in rotating shift workers on a night shift schedule. To determine predicted circadian phase, center of gravity of the fitted bimodal skewed baseline cosine curve was used for melatonin, and acrophase of the cosine curve for aMT6s. On a fixed sleep schedule, the model predicted melatonin phase to within ± 1 hour in 67% and ± 1.5 hours in 100% of participants, with mean absolute error of 41 ± 32 minutes. On diurnal schedules, including shift workers, the model predicted aMT6s acrophase to within ± 1 hour in 66% and ± 2 hours in 87% of participants, with mean absolute error of 63 ± 67 minutes. On night shift schedules, the model predicted aMT6s acrophase to within ± 1 hour in 42% and ± 2 hours in 53% of participants, with mean absolute error of 143 ± 155 minutes. Prediction accuracy was similar when using either 1 (wrist) or 11 skin temperature sensor inputs. These findings demonstrate that the model can predict circadian timing to within ± 2 hours for the vast majority of individuals on diurnal schedules, using blue light and a single temperature sensor. However, this approach did not generalize to night shift conditions.
| S-EPMC6662750 | biostudies-literature
Project description:Intracranial hemorrhage is a life-threatening condition, the outcome of which can be improved by intensive care. Intracranial hemorrhage may be spontaneous, precipitated by an underlying vascular malformation, induced by trauma, or related to therapeutic anticoagulation. The goals of critical care are to assess the proximate cause, minimize the risks of hemorrhage expansion through blood pressure control and correction of coagulopathy, and obliterate vascular lesions with a high risk of acute rebleeding. Simple bedside scales and interpretation of computed tomography scans assess the severity of neurological injury. Myocardial stunning and pulmonary edema related to neurological injury should be anticipated, and can usually be managed. Fever (often not from infection) is common and can be effectively treated, although therapeutic cooling has not been shown to improve outcomes after intracranial hemorrhage. Most functional and cognitive recovery takes place weeks to months after discharge; expected levels of functional independence (no disability, disability but independence with a device, dependence) may guide conversations with patient representatives. Goals of care impact mortality, with do-not-resuscitate status increasing the predicted mortality for any level of severity of intraparenchymal hemorrhage. Future directions include refining the use of bedside neuro-monitoring (electroencephalogram, invasive monitors), novel approaches to reduce intracranial hemorrhage expansion, minimizing vasospasm, and refining the assessment of quality of life to guide rehabilitation and therapy.
| S-EPMC3361326 | biostudies-literature
Project description:Intracranial hemorrhage refers to any bleeding within the intracranial vault, including the brain parenchyma and surrounding meningeal spaces. This article focuses on the acute diagnosis and management of primary nontraumatic intracerebral hemorrhage and subarachnoid hemorrhage in the emergency department.
| S-EPMC3443867 | biostudies-literature
Project description:Organisations and governments seeking to implement genomics into clinical practice face numerous challenges across multiple, diverse aspects of the health care system. It is not sufficient to tackle any one aspect in isolation: to create a system that supports genomic medicine, they must be addressed simultaneously. The growing body of global knowledge can guide decision-making, but each jurisdiction or organisation needs a model for genomic (or personalised) medicine that is tailored to its unique context, its priorities and the funds available. Poor decisions could greatly reduce the benefits that could potentially arise from genomic medicine. Demonstration projects enable models to be tested, providing valuable evidence and experience for subsequent implementation. Here, we present the Melbourne Genomics Health Alliance demonstration project as an exemplar of a collaborative, holistic approach to phased implementation of genomics across multiple autonomous institutions. The approach and lessons learned may assist others in determining how best to integrate genomics into their healthcare system.
| S-EPMC5677913 | biostudies-literature
Project description:BackgroundPrediction tools for various intraoperative bleeding events remain scarce. We aim to develop machine learning-based models and identify the most important predictors by real-world data from electronic medical records (EMRs).MethodsAn established database of surgical inpatients in Shanghai was utilized for analysis. A total of 51,173 inpatients were assessed for eligibility. 48,543 inpatients were obtained in the dataset and patients were divided into haemorrhage (N = 9728) and without-haemorrhage (N = 38,815) groups according to their bleeding during the procedure. Candidate predictors were selected from 27 variables, including sex (N = 48,543), age (N = 48,543), BMI (N = 48,543), renal disease (N = 26), heart disease (N = 1309), hypertension (N = 9579), diabetes (N = 4165), coagulopathy (N = 47), and other features. The models were constructed by 7 machine learning algorithms, i.e., light gradient boosting (LGB), extreme gradient boosting (XGB), cathepsin B (CatB), Ada-boosting of decision tree (AdaB), logistic regression (LR), long short-term memory (LSTM), and multilayer perception (MLP). An area under the receiver operating characteristic curve (AUC) was used to evaluate the model performance.ResultsThe mean age of the inpatients was 53 ± 17 years, and 57.5% were male. LGB showed the best predictive performance for intraoperative bleeding combining multiple indicators (AUC = 0.933, sensitivity = 0.87, specificity = 0.85, accuracy = 0.87) compared with XGB, CatB, AdaB, LR, MLP and LSTM. The three most important predictors identified by LGB were operative time, D-dimer (DD), and age.ConclusionsWe proposed LGB as the best Gradient Boosting Decision Tree (GBDT) algorithm for the evaluation of intraoperative bleeding. It is considered a simple and useful tool for predicting intraoperative bleeding in clinical settings. Operative time, DD, and age should receive attention.
| S-EPMC10416513 | biostudies-literature
Project description:BackgroundRapid magnetic resonance imaging (MRI) protocols may be effective in the emergency department (ED) to evaluate nontraumatic neurologic complaints. We evaluate neuroimaging (rapid MRI [rMRI]), head computerized tomography [HCT], and full MRI) use following widespread implementation of rMRI protocols in a pediatric emergency department (ED).MethodsWe conducted a retrospective study in a tertiary care pediatric ED of encounters with neuroimaging during two 9-month periods: one prior to (control period) and one after generalized availability of 4 rMRI protocols (rMRI period). The primary outcome was differences in neuroimaging rates between the two periods. Secondary outcomes included ED process measures, unsuccessful imaging, and undetected pathology, with full MRI within 14?days as the reference standard.ResultsThere were 1052 encounters with neuroimaging during the control and 1308 during the rMRI periods. Differences in neuroimaging between periods were 27.7% for rMRI (95% CI, 24.4, 31.0), -?21.5% for HCT (95% CI, -?25.5, -?17.5), and?-?6.2% for full MRI (95% CI, -?9.3, -?3.1%.) Time to imaging (182 [IQR 138-255] versus 86 [IQR 52-137] minutes) as well as ED length of stay (396 [IQR 304-484] versus 257 [IQR 196-334] minutes) was longer for rMRI versus HCT (p?<?0.01). Between the control and rMRI periods, there were differences in types of neuroimaging performed for patients with altered mental status, headache, seizure, shunt dysfunction, stroke, syncope, trauma, vomiting, infection, and other neurologic complaints (p?<?0.05). rMRI studies were unsuccessful in 3.6% of studies versus 0.0% of HCTs (p?<?0.01). The 22 unsuccessful rMRI studies were unsuccessful due to artifacts from dental hardware (n?=?2) and patient motion (n?=?20). None of the rMRI studies with full MRI follow-up imaging had undetected pathology; the false negative rate for the HCT exams was as high as 25%.ConclusionsAfter routine ED use of 4 rMRI protocols, there was a more than 20% decrease in HCT use without missed diagnoses. Time to neuroimaging and length of stay were longer for rMRI than HCT, with higher rates of unsuccessful imaging. Despite these limitations, rMRI may be an alternative to HCT for nontraumatic complaints in the ED.
| S-EPMC6956479 | biostudies-literature
Project description:ImportanceAn adequate system for triaging patients with head trauma in prehospital settings and choosing optimal medical institutions is essential for improving the prognosis of these patients. To our knowledge, there has been no established way to stratify these patients based on their head trauma severity that can be used by ambulance crews at an injury site.ObjectivesTo develop a prehospital triage system to stratify patients with head trauma according to trauma severity by using several machine learning techniques and to evaluate the predictive accuracy of these techniques.Design, setting, and participantsThis single-center retrospective cohort study was conducted by reviewing the electronic medical records of consecutive patients who were transported to Tokyo Medical and Dental University Hospital in Japan from April 1, 2018, to March 31, 2021. Patients younger than 16 years with cardiopulmonary arrest on arrival or with a significant amount of missing data were excluded.Main outcomes and measuresMachine learning-based prediction models to detect the presence of traumatic intracranial hemorrhage were constructed. The predictive accuracy of the models was evaluated with the area under the receiver operating curve (ROC-AUC), area under the precision recall curve (PR-AUC), sensitivity, specificity, and other representative statistics.ResultsA total of 2123 patients (1527 male patients [71.9%]; mean [SD] age, 57.6 [19.8] years) with head trauma were enrolled in this study. Traumatic intracranial hemorrhage was detected in 258 patients (12.2%). Among several machine learning algorithms, extreme gradient boosting (XGBoost) achieved the mean (SD) highest ROC-AUC (0.78 [0.02]) and PR-AUC (0.46 [0.01]) in cross-validation studies. In the testing set, the ROC-AUC was 0.80, the sensitivity was 74.0% (95% CI, 59.7%-85.4%), and the specificity was 74.9% (95% CI, 70.2%-79.3%). The prediction model using the National Institute for Health and Care Excellence (NICE) guidelines, which was calculated after consultation with physicians, had a sensitivity of 72.0% (95% CI, 57.5%-83.8%) and a specificity of 73.3% (95% CI, 68.7%-77.7%). The McNemar test revealed no statistically significant differences between the XGBoost algorithm and the NICE guidelines for sensitivity or specificity (P = .80 and P = .55, respectively).Conclusions and relevanceIn this cohort study, the prediction model achieved a comparatively accurate performance in detecting traumatic intracranial hemorrhage using only the simple pretransportation information from the patient. Further validation with a prospective multicenter data set is needed.
| S-EPMC9187955 | biostudies-literature
Project description:Background and purposeIdentification of ICH, particularly after ischemic stroke therapy, is important for guiding subsequent antithrombotic management and is often confounded by contrast staining or extravasations within intracerebral or extra-axial compartments. This study evaluates the accuracy of DECT in distinguishing ICH from iodinated contrast in patients who received contrast via IA or IV delivery.Materials and methodsForty patients who had received IA or IV contrast were evaluated using a DECT scanner at 80kV and 140kV to distinguish hyperdensities secondary to contrast staining or extravasation from those representing ICH. A 3-material decomposition algorithm was used to obtain virtual noncontrast images and iodine overlay images. Sensitivity, specificity, and accuracy of DECT in prospectively distinguishing intracranial contrast from hemorrhage within parenchymal, subarachnoid, extra-axial, intraventricular, and intra-arterial compartments were computed using routine clinical follow-up imaging as the standard of reference.ResultsA total of 148 foci of intracranial hyperattenuation were identified. Of these, 142 were correctly classified for the presence of hemorrhage by DECT. The sensitivity, specificity, and accuracy for identifying hemorrhage, depending on the compartment being considered, were 100%, 84.4%-100%, and 87.2%-100%, respectively. The only instances where DECT failed to correctly identify the source of hyperattenuation was in the presence of diffuse parenchymal calcification (n = 5) and a metallic streak artifact (n = 1).ConclusionAfter IA and/or IV contrast administration, DECT can accurately differentiate all types of ICH from iodinated contrast without employing any additional radiation.
| S-EPMC8013231 | biostudies-literature
Project description:PurposeIn our conventional image registration workflow, the four-dimensional (4D) CBCT was directly registered to the reference helical CT (HCT) using a dual registration approach within the Elekta XVI software. In this study, we proposed a new HCT-CBCT auto-registration strategy using a previously registered CBCT (CBCTpre) as the reference image and tested its clinical feasibility.MethodsFrom a previous CBCT session, the registered average 4D CBCT was selected as CBCTpre and the HCT-CBCTpre registration vector from the clinician's manual registration result was recorded. In the new CBCT session, auto-registration was performed between the new average 4D CBCT (CBCTtx) and CBCTpre (CBCTpre-CBCTtx). The overall HCT-CBCTtx registration result was then derived by combing the results from two registrations (i.e., HCT-CBCTpre + CBCTpre-CBCTtx). The results from the proposed method were compared with clinician's manually adjusted HCT-CBCTtx registration results ("ground truth") to evaluate its accuracy using a test dataset consisting of 32 challenging registration cases.ResultsThe uncertainty of the proposed auto-registration method was -0.1 ± 0.5, 0.1 ± 1.0, and -0.1 ± 0.7 mm in three translational directions (lateral, longitudinal, and vertical) and 0.0° ± 0.9°, 0.3° ± 0.9°, and 0.4° ± 0.7° in three rotation directions, respectively. Two patients (6.3%) had translational uncertainty > 2 mm (max = 3.1 mm) and both occurred in the longitudinal direction. Meanwhile, the uncertainty of the conventional direct HCT-CBCTtx auto-registration was -0.4 ± 2.6, -0.2 ± 7.4, -1.4 ± 3.6 mm for translations and -0.3° ± 1.2°, 0.0° ± 1.6°, and 0.1 ± 1.1° for rotations. Eleven patients (34.4%) had translation uncertainty > 2 mm (max = 26.2 mm) in at least one direction. Accuracy in translation was improved with the new method, while rotation accuracy stayed in the same order.ConclusionWe demonstrated the feasibility of incorporating prior clinical registration knowledge into the online HCT-CBCT registration process. The proposed auto-registration method provides a quick and reliable starting solution for online HCT-CBCT registration.
| S-EPMC8992944 | biostudies-literature