EEG Tensorization Enhances CNN-Based Outcome Classification in Comatose Patients Following a Cardiac Arrest.

Funding Acknowledgements Type of funding sources: None. Introduction Out-of-hospital cardiac arrest (OHCA) is a major public health challenge and although rate of intrahospital survival increased over the last 40 years, it still remains poor (from 8,6% in 1976-1999 to 9,9% in 2000-2019). Different studies report that introduction of mild therapeutic hypothermia (TTM) improves survival and neurological outcome in comatose patients after OHCA. Purpose The aim of this research was to evaluate influence of pre-hospital predictors related to cardiopulmonary resuscitation (CPR), neurological status and ECG changes at admission and early percutaneous coronary intervention (PCI) performed within 24h of admission on intrahospital survival and neurological outcome of OHCA patients. Methods The research was conducted as a retrospective cohort study of data taken from the hospital registry on OHCA from January 2007 until November 2019. The analyzed factors were: bystander CPR, duration of CPR until return of ROSC, initial rhythm, responsiveness upon admission defined as Glasgow Coma Score (GCS)>8, presence of ST segment elevation (STEMI) on electrocardiography (ECG) and early PCI. The favorable neurological outcome was defined as a cerebral performance category scale (CPC)≤2. Results The research included 506 survivors of OHCA. Cardiac arrest was witnessed in 412 (81.4%), bystander CPR was performed in 197 (38.9%), CPR lasted ≤20min in 291 (57.5%), initial rhythm was shockable in 304 (60.1%) of patients. At admission 387 (76.5%) were comatose (GCS < 8) and TTM was introduced in 177 (45.7%) of patients. ECG upon admission showed STEMI in 176 (34.8%) and early PCI was performed in 145 (28.6%) of patients. In-hospital mortality in our study group was 281 (55.5%) and 185 (36.6%) of patients had favorable neurological outcome. Multivariate regression analysis showed that initial shockable rhythm (OR 3.391 [2.310-4.977], p < 0.0005), early PCI (OR 0.368 [0.226-0.599], p < 0.0005), duration of CPR ≤20min (OR 4.249 [2.688-6.718], p < 0.0005) and GCS > 8 (OR 0.194 [0.110-0.343], p < 0.0005) were independent predictors of in-hospital survival. Independent predictors of favorable neurological outcome were: initial shockable rhythm (OR 3.301 [2.002-5.441], p< 0.0005), STEMI on ECG upon admission (OR 0.528 [0.326-0.853], p = 0.009), duration of CPR ≤20min (OR 5.144 [3.090-8.565], p< 0.0005) and GCS > 8 (OR 0.152 [0.088-0.260], p< 0.0005). Introduction of TTM improved both intrahospital survival (54.1% vs. 24.4%; p < 0.0005) and neurological outcome (33.5% vs. 11.6%; p < 0.0005) in patients with initial shockable rhythm. Conclusion In our study group of OHCA patients of any origin, initial shockable rhythm, duration of CPR ≤20min and GCS > 8 at admission influenced both intrahospital survival and favorable neurological outcome. Introduction of TTM significantly improved both survival and neurological outcome in comatose patients with initial shockable rhythm.

Introduction/Objective: Survival and neurologic recovery after out-of-hospital cardiac arrest remain poor despite significant advances in the therapeutic approach. The study aimed to evaluate predictors of intrahospital survival and neurologic outcome among patients after outof-hospital cardiac arrest as well as to evaluate the influence of mild therapeutic hypothermia introduction on intrahospital survival and neurologic outcome among comatose patients after out-of-hospital cardiac arrest. Methods The research was conducted as a retrospective observational study among patients hospitalized at the Cardiac Intensive Care Unit of the Institute for Cardiovascular Diseases of Vojvodina from January 2007 until November 2019 as a result of an out-of-hospital cardiac arrest. Results. The research included 506 survivors of OHCA. Multivariate regression analysis showed that initial shockable rhythm, cardiopulmonary resuscitation efforts lasting no longer than 20 minutes and a Glasgow Coma Score above 8 at admission, were predictors of intrahospital survival and good neurological outcome. Introduction of mild therapeutic hypothermia improved intrahospital survival (54.1% vs. 24.4%; p < 0.0005) and neurological outcome (42.9% vs. 18.3%; p < 0.0005) in comatose patients with initial shockable rhythm. Conclusion. In our study group of out-of-hospital cardiac arrest patients, initial shockable rhythm, cardiopulmonary resuscitation efforts lasting no longer than 20min and a Glasgow Coma Score above 8 at admission were predictors of intrahospital survival and favourable neurological outcome. The introduction of mild therapeutic hypothermia significantly improved survival and neurological outcomes in comatose patients with initial shockable rhythms.

Stacking ensemble learning models for daily runoff prediction using 1D and 2D CNNs

This study used a retrospective data analysis approach combined with model development and validation. The present study introduces a 2.5D convolutional neural network (CNN) model leveraging CT imaging to facilitate the early detection of malignant vertebral compression fractures (MVCFs), potentially reducing reliance on invasive biopsies. Vertebral histopathologic biopsy is recognized as the gold standard for differentiating between osteoporotic and malignant vertebral compression fractures (VCFs). Nevertheless, its application is restricted due to its invasive nature and high cost, highlighting the necessity for alternative methods to identify MVCFs. The clinical, imaging, and pathologic data of patients who underwent vertebral augmentation and biopsy at institution 1 and institution 2 were collected and analyzed. On the basis of the vertebral CT images of these patients, 2D, 2.5D, and 3D CNN models were developed to identify the patients with osteoporotic vertebral compression fractures (OVCF) and MVCF. To verify the clinical application value of the CNN model, two rounds of reader studies were performed. The 2.5D CNN model performed well, and its performance in identifying MVCF patients was significantly superior to that of the 2D and 3D CNN models. In the training data set, the area under the receiver operating characteristic curve (AUC) of the 2.5D CNN model was 0.996 and an F1 score of 0.915. In the external cohort test, the AUC was 0.815 and an F1 score of 0.714. And clinicians' ability to identify MVCF patients has been enhanced by the 2.5D CNN model. With the assistance of the 2.5D CNN model, the AUC of senior clinicians was 0.882, and the F1 score was 0.774. For junior clinicians, the 2.5D CNN model-assisted AUC was 0.784 and the F1 score was 0.667. The development of our 2.5D CNN model marks a significant step toward noninvasive identification of MVCF patients. The 2.5D CNN model may be a potential model to assist clinicians in better identifying MVCF patients.

A CONVOLUTIONAL NEURAL NETWORK FOR AUTOMATIC DISCRIMINATION OF PAUSE EPISODES DETECTED BY AN INSERTABLE CARDIAC MONITOR

10. Early EEG/SEPs recordings are reliable predictors of good and poor outcome in comatose patients after cardiac arrest

Anecdotal reports suggest that a loss of distinction between gray (GM) and white matter (WM) as adjudged by CT scan predicts poor outcome in comatose patients after cardiac arrest. To address this, we quantitatively assessed GM and WM intensities at various brain levels in comatose patients after cardiac arrest. Patients for whom consultation was requested within 24 hours of a cardiac arrest were identified with the use of a computerized database that tracks neurological consultations at our institution. Twenty-five comatose patients were identified for whom complete medical records and CT scans were available for review. Twenty-five consecutive patients for whom a CT scan was interpreted as normal served as controls. Hounsfield units (HUs) were measured in small defined areas obtained from axial images at the levels of the basal ganglia, centrum semiovale, and high convexity area. At each level tested, lower GM intensity and higher WM intensity were noted in comatose patients compared with normal controls. The GM/WM ratio was significantly lower among comatose patients compared with controls (P:<0.0001, rank sum test). There was essentially no overlap in GM/WM ratios between control and study patients. The difference was greatest at the basal ganglia level. We also observed a marginally significant difference in the GM/WM ratio at the basal ganglia level between those patients who died and those who survived cardiac arrest (P:=0. 035, 1-tailed t test). Using receiver operating characteristic curve analysis, we determined that a difference in GM/WM ratio of <1.18 at the basal ganglia level was 100% predictive of death. At the basal ganglia level, none of 12 patients below this threshold survived, whereas the survival rate was 46% among patients in whom the ratio was >1.18. The empirical risk of death was 21.67 for comatose patients with a value below threshold. The ratio in HUs of GM to WM provides a reproducible measure of the distinction between gray and white matter. A lower GM/WM ratio is observed in comatose patients immediately after cardiac arrest. The basal ganglia level seems to be the most sensitive location on CT for measuring this relationship. Although a GM/WM ratio <1.18 at this level predicted death in this retrospective study, the difference in this study is not robust enough to recommend that management decisions be dictated by CT results. The results, however, do warrant consideration of a prospective study to determine the reliability of CT scanning in predicting outcome for comatose patients after cardiac arrest.

Deep learning models have been successfully applied to the analysis of various functional MRI data. Convolutional neural networks (CNN), a class of deep neural networks, have been found to excel at extracting local meaningful features based on their shared-weights architecture and space invariance characteristics. In this study, we propose M2D CNN, a novel multichannel 2D CNN model, to classify 3D fMRI data. The model uses sliced 2D fMRI data as input and integrates multichannel information learned from 2D CNN networks. We experimentally compared the proposed M2D CNN against several widely used models including SVM, 1D CNN, 2D CNN, 3D CNN, and 3D separable CNN with respect to their performance in classifying task-based fMRI data. We tested M2D CNN against six models as benchmarks to classify a large number of time-series whole-brain imaging data based on a motor task in the Human Connectome Project (HCP). The results of our experiments demonstrate the following: (i) convolution operations in the CNN models are advantageous for high-dimensional whole-brain imaging data classification, as all CNN models outperform SVM; (ii) 3D CNN models achieve higher accuracy than 2D CNN and 1D CNN model, but 3D CNN models are computationally costly as any extra dimension is added in the input; (iii) the M2D CNN model proposed in this study achieves the highest accuracy and alleviates data overfitting given its smaller number of parameters as compared with 3D CNN.

Identification of slightly sprouted wheat kernels using hyperspectral imaging technology and different deep convolutional neural networks

Purpose: Detection of Alzheimer's disease (AD) on magnetic resonance imaging (MRI) using convolutional neural networks (CNNs), which is useful for detecting AD in its preliminary states. Approach: Our study implements and compares several deep models and configurations, including two-dimensional (2D) and three-dimensional (3D) CNNs and recurrent neural networks (RNNs). To use a 2D CNN on 3D MRI volumes, each MRI scan is split into 2D slices, neglecting the connection among 2D image slices in an MRI volume. Instead, a CNN model could be followed by an RNN in a way that the model of 2D CNN + RNN can understand the connection among sequences of 2D image slices for an MRI. The issue is that the feature extraction step in the 2D CNN is independent of classification in the RNN. To tackle this, 3D CNNs can be employed instead of 2D CNNs to make voxel-based decisions. Our study's main contribution is to introduce transfer learning from a dataset of 2D images to 3D CNNs. Results: The results on our MRI dataset indicate that sequence-based decisions improve the accuracy of slice-based decisions by 2% in classifying AD patients from healthy subjects. Also the 3D voxel-based method with transfer learning outperforms the other methods with 96.88% accuracy, 100% sensitivity, and 94.12% specificity. Conclusions: Several implementations and experiments using CNNs on MRI scans for AD detection demonstrated that the voxel-based method with transfer learning from ImageNet to MRI datasets using 3D CNNs considerably improved the results compared with the others.

Survey on current practices for neurological prognostication after cardiac arrest

To validate quantitative diffusion-weighted imaging (DWI) MRI thresholds that correlate with poor outcome in comatose cardiac arrest survivors, we conducted a clinician-blinded study and prospectively obtained MRIs from comatose patients after cardiac arrest. Consecutive comatose post-cardiac arrest adult patients were prospectively enrolled. MRIs obtained within 7 days after arrest were evaluated. The clinical team was blinded to the DWI MRI results and followed a prescribed prognostication algorithm. Apparent diffusion coefficient (ADC) values and thresholds differentiating good and poor outcome were analyzed. Poor outcome was defined as a Glasgow Outcome Scale score of ≤2 at 6 months after arrest. Ninety-seven patients were included, and 75 patients (77%) had MRIs. In 51 patients with MRI completed by postarrest day 7, the prespecified threshold of >10% of brain tissue with an ADC <650 ×10-6 mm2/s was highly predictive for poor outcome with a sensitivity of 0.63 (95% confidence interval [CI] 0.42-0.80), a specificity of 0.96 (95% CI 0.77-0.998), and a positive predictive value (PPV) of 0.94 (95% CI 0.71-0.997). The mean whole-brain ADC was higher among patients with good outcomes. Receiver operating characteristic curve analysis showed that ADC <650 ×10-6 mm2/s had an area under the curve of 0.79 (95% CI 0.65-0.93, p < 0.001). Quantitative DWI MRI data improved prognostication of both good and poor outcomes. This prospective, clinician-blinded study validates previous research showing that an ADC <650 ×10-6 mm2/s in >10% of brain tissue in an MRI obtained by postarrest day 7 is highly specific for poor outcome in comatose patients after cardiac arrest.

10. Prediction of good and poor outcome in comatose patients after cardiac arrest: The utility of early EEG/SEP recordings during therapeutic hypothermia

Objective There is no specific treatment for comatose patients after near-hanging or in those who recover from cardiac arrest (CA) caused by hanging. Since 2009, we have used therapeutic hypothermia (TH) to treat all comatose survivors of near-hanging and in patients who recovered from CA caused by hanging. The purpose of this study was to describe the outcomes in comatose patients after near-hanging. Design Case series. Setting Emergency departments of two regional hospitals. Methods We collected patient data from the Samsung Medical Center hypothermia database between November 2009 and November 2011. We included all patients presented with near-hanging or CA caused by hanging; who remained comatose and received TH after resuscitation for analysis. Clinical characteristics and outcome of patients were presented. Results During the study period, 26 patients were admitted to the emergency department after near-hanging or CA caused by hanging; 21 patients were enrolled in this study. Twelve patients with CA and 9 comatose patients without CA were treated with TH. Only 1 patient with CA had a good neurological outcome. By contrast, all near-hanging patients without CA had a good neurological outcome. Conclusions TH can be an effective therapeutic modality in cases of near-hanging without CA. However, the effectiveness of TH is questionable in patients who survive from CA caused by hanging. (Hong Kong j.emerg.med. 2014;21:316-321)

The improved soft tissue contrast of magnetic resonance imaging(MRI) compared to computed tomography (CT) makes it a useful imaging modality for radiotherapy treatment planning. Even when MR images are acquired for treatment planning, the standard clinical practice currently also requires a CT for dose calculation and x-ray-based patient positioning. This increases workloads, introduces uncertainty due to the required inter-modality image registrations, and involves unnecessary irradiation. While it would be beneficial to use exclusively MR images, a method needs to be employed to estimate a synthetic CT (sCT) for generating electron density maps and patient positioning reference images. We investigated 2D and 3D convolutional neural networks (CNNs) to generate a male pelvic sCT using a T1-weighted MR image and compare their performance. A retrospective study was performed using CTs and T1-weighted MR images of 20 prostate cancer patients. CTs were deformably registered to MR images to create CT-MR pairs for training networks. The proposed 2D CNN, which contained 27 convolutional layers, was modified from the state-of-the-art2D CNN to save computational memory and prepare for building the 3D CNN. The proposed 2D and 3D models were trained from scratch to map intensities of T1-weighted MR images to CT Hounsfield Unit (HU) values. Each sCT was generated in a fivefold cross-validation framework and compared with the corresponding deformed CT (dCT) using voxel-wise mean absolute error (MAE). The sCT geometric accuracy was evaluated by comparing bone regions, defined by thresholding at 150HU in the dCTs and the sCTs, using dice similarity coefficient (DSC), recall, and precision. To evaluate sCT patient positioning accuracy, bone regions in dCTs and sCTs were rigidly registered to the corresponding cone-beam CTs. The resulting paired Euler transformation vectors were compared by calculating translation vector distances and absolute differences of Euler angles. Statistical tests were performed to evaluate the differences among the proposed models and Han's model. Generating a pelvic sCT required approximately 5.5s using the proposed models. The average MAEs within the body contour were 40.5±5.4HU (mean±SD) and 37.6±5.1HU for the 2D and 3D CNNs, respectively. The average DSC, recall, and precision for the bone region (thresholding the CT at 150HU) were 0.81±0.04, 0.85±0.04, and 0.77±0.09 for the 2D CNN, and 0.82±0.04, 0.84±0.04, and 0.80±0.08 for the 3D CNN, respectively. For both models, mean translation vector distances are less than 0.6mm with mean absolute differences of Euler angles less than 0.5°. The 2D and 3D CNNs generated accurate pelvic sCTs for the 20 patients using T1-weighted MR images. Statistical tests indicated that the proposed 3D model was able to generate sCTs with smaller MAE and higher bone region precision compared to 2D models. Results of patient alignment tests suggested that sCTs generated by the proposed CNNs can provide accurate patient positioning. The accuracy of the dose calculation using generated sCTs will be tested and compared for the proposed models in the future.