The medical imaging literature has witnessed remarkable progress in high-performing segmentation models based on convolutional neural networks. Despite the new performance highs, the recent advanced segmentation models still require large, representative, and high quality annotated datasets. However, rarely do we have a perfect training dataset, particularly in the field of medical imaging, where data and annotations are both expensive to acquire. Recently, a large body of research has studied the problem of medical image segmentation with imperfect datasets, tackling two major dataset limitations: scarce annotations where only limited annotated data is available for training, and weak annotations where the training data has only sparse annotations, noisy annotations, or image-level annotations. In this article, we provide a detailed review of the solutions above, summarizing both the technical novelties and empirical results. We further compare the benefits and requirements of the surveyed methodologies and provide our recommended solutions to the problems of scarce and weak annotations. We hope this review increases the community awareness of the techniques to handle imperfect datasets.

Objective: What mechanisms underlie the loss and recovery of consciousness after severe brain injury? We sought to establish, in the largest cohort of patients with disorders of consciousness (DOC) to date, the link between gold standard clinical measures of awareness and wakefulness, and specific patterns of local brain pathology-thereby possibly providing a mechanistic framework for patient diagnosis, prognosis, and treatment development. Methods: Structural T1-weighted magnetic resonance images were collected, in a continuous sample of 143 severely brain-injured patients with DOC (and 96 volunteers), across 2 tertiary expert centers. Brain atrophy in subcortical regions (bilateral thalamus, basal ganglia, hippocampus, basal forebrain, and brainstem) was assessed across (1) healthy volunteers and patients, (2) clinical entities (eg, vegetative state, minimally conscious state), (3) clinical measures of consciousness (Coma Recovery Scale-Revised), and (4) injury etiology. Results: Compared to volunteers, patients exhibited significant atrophy across all structures (p < 0.05, corrected). Strikingly, we found almost no significant differences across clinical entities. Nonetheless, the clinical measures of awareness and wakefulness upon which differential diagnosis rely were systematically associated with tissue atrophy within thalamic and basal ganglia nuclei, respectively; the basal forebrain was atrophied in proportion to patients' response to sensory stimulation. In addition, nontraumatic injuries exhibited more extensive thalamic atrophy. Interpretation: These findings provide, for the first time, a grounding in pathology for gold standard behavior-based clinical measures of consciousness, and reframe our current models of DOC by stressing the different links tying thalamic mechanisms to willful behavior and extrathalamic mechanisms to behavioral (and electrocortical) arousal.ANN NEUROL 2015;00:000-000 T he mechanisms supporting consciousness, as well as its loss and recovery after severe brain injury, remain largely unknown. In the context of disorders of consciousness (DOC) 1 such as the vegetative state (VS) and the minimally conscious state (MCS), the lack of a mechanistic understanding of the relationship between brain damage and neurological condition has direct consequences for our ability to make accurate diagnoses, prognoses, and to develop targeted interventions, thereby raising complicated medical and ethical questions. 2Although information concerning the nature and extent of a patient's brain damage is generally taken into consideration during clinical assessments, current differential diagnosis procedures rely exclusively-as per international guidelines-on behavioral presentation. 3-5 Consequently, although our understanding of DOC is continuously increasing, 6,7 little is known about the connection between behaviorally defined clinical entities and the underlying brain damage, [8][9][10] or the degree to which standard behavior-based clinical assessments (eg, JFK View this article online at wil...

The study of structural and functional magnetic resonance imaging data has greatly benefitted from the development of sophisticated and efficient algorithms aimed at automating and optimizing the analysis of brain data. We address, in the context of the segmentation of brain from non-brain tissue (i.e., brain extraction, also known as skull-stripping), the tension between the increased theoretical and clinical interest in patient data, and the difficulty of conventional algorithms to function optimally in the presence of gross brain pathology. Indeed, because of the reliance of many algorithms on priors derived from healthy volunteers, images with gross pathology can severely affect their ability to correctly trace the boundaries between brain and non-brain tissue, potentially biasing subsequent analysis. We describe and make available an optimized brain extraction script for the pathological brain (optiBET) robust to the presence of pathology. Rather than attempting to trace the boundary between tissues, optiBET performs brain extraction by (i) calculating an initial approximate brain extraction; (ii) employing linear and non-linear registration to project the approximate extraction into the MNI template space; (iii) back-projecting a standard brain-only mask from template space to the subject’s original space; and (iv) employing the back-projected brain-only mask to mask-out non-brain tissue. The script results in up to 94% improvement of the quality of extractions over those obtained with conventional software across a large set of severely pathological brains. Since optiBET makes use of freely available algorithms included in FSL, it should be readily employable by anyone having access to such tools.

With the continuing coronavirus disease 2019 (COVID-19) pandemic coupled with phased reopening, it is critical to identify risk factors associated with susceptibility and severity of disease in a diverse population to help shape government policies, guide clinical decision making, and prioritize future COVID-19 research. In this retrospective case-control study, we used de-identified electronic health records (EHR) from the University of California Los Angeles (UCLA) Health System between March 9th, 2020 and June 14th, 2020 to identify risk factors for COVID-19 susceptibility (severe acute respiratory distress syndrome coronavirus 2 (SARS-CoV-2) PCR test positive), inpatient admission, and severe outcomes (treatment in an intensive care unit or intubation). Of the 26,602 individuals tested by PCR for SARS-CoV-2, 992 were COVID-19 positive (3.7% of Tested), 220 were admitted in the hospital (22% of COVID-19 positive), and 77 had a severe outcome (35% of Inpatient). Consistent with previous studies, males and individuals older than 65 years old had increased risk of inpatient admission. Notably, individuals self-identifying as Hispanic or Latino constituted an increasing percentage of COVID-19 patients as disease severity escalated, comprising 24% of those testing positive, but 40% of those with a severe outcome, a disparity that remained after correcting for medical co-morbidities. Cardiovascular disease, hypertension, and renal disease were premorbid risk factors present before SARS-CoV-2 PCR testing associated with COVID-19 susceptibility. Less well-established risk factors for COVID-19 susceptibility included pre-existing dementia (odds ratio (OR) 5.2 [3.2-8.3], p=2.6 x 10-10), mental health conditions (depression OR 2.1 [1.6-2.8], p=1.1 x 10-6) and vitamin D deficiency (OR 1.8 [1.4-2.2], p=5.7 x 10-6). Renal diseases including end-stage renal disease and anemia due to chronic renal disease were the predominant premorbid risk factors for COVID-19 inpatient admission. Other less established risk factors for COVID-19 inpatient admission included previous renal transplant (OR 9.7 [2.8-39], p=3.2x10-4) and disorders of the immune system (OR 6.0 [2.3, 16], p=2.7x10-4). Prior use of oral steroid medications was associated with decreased COVID-19 positive testing risk (OR 0.61 [0.45, 0.81], p=4.3x10-4), but increased inpatient admission risk (OR 4.5 [2.3, 8.9], p=1.8x10-5). We did not observe that prior use of angiotensin converting enzyme inhibitors or angiotensin receptor blockers increased the risk of testing positive for SARS-CoV-2, being admitted to the hospital, or having a severe outcome. This study involving direct EHR extraction identified known and less well-established demographics, and prior diagnoses and medications as risk factors for COVID-19 susceptibility and inpatient admission. Knowledge of these risk factors including marked ethnic disparities observed in disease severity should guide government policies, identify at-risk populations, inform clinical decision making, and prioritize future COVID-19 research.

Worldwide, testing capacity for SARS-CoV-2 is limited and bottlenecks in the scale up of polymerase chain reaction (PCR-based testing exist. Our aim was to develop and evaluate a machine learning algorithm to diagnose COVID-19 in the inpatient setting. The algorithm was based on basic demographic and laboratory features to serve as a screening tool at hospitals where testing is scarce or unavailable. We used retrospectively collected data from the UCLA Health System in Los Angeles, California. We included all emergency room or inpatient cases receiving SARS-CoV-2 PCR testing who also had a set of ancillary laboratory features (n = 1,455) between 1 March 2020 and 24 May 2020. We tested seven machine learning models and used a combination of those models for the final diagnostic classification. In the test set (n = 392), our combined model had an area under the receiver operator curve of 0.91 (95% confidence interval 0.87-0.96). The model achieved a sensitivity of 0.93 (95% CI 0.85-0.98), specificity of 0.64 (95% CI 0.58-0.69). We found that our machine learning algorithm had excellent diagnostic metrics compared to SARS-CoV-2 PCR. This ensemble machine learning algorithm to diagnose COVID-19 has the potential to be used as a screening tool in hospital settings where PCR testing is scarce or unavailable.

In two-thirds of intensive care unit (ICU) patients and 90% of surgical patients, arterial blood pressure (ABP) is monitored non-invasively but intermittently using a blood pressure cuff. Since even a few minutes of hypotension increases the risk of mortality and morbidity, for the remaining (high-risk) patients ABP is measured continuously using invasive devices, and derived values are extracted from the recorded waveforms. However, since invasive monitoring is associated with major complications (infection, bleeding, thrombosis), the ideal ABP monitor should be both non-invasive and continuous. With large volumes of high-fidelity physiological waveforms, it may be possible today to impute a physiological waveform from other available signals. Currently, the state-of-the-art approaches for ABP imputation only aim at intermittent systolic and diastolic blood pressure imputation, and there is no method that imputes the continuous ABP waveform. Here, we developed a novel approach to impute the continuous ABP waveform non-invasively using two continuously-monitored waveforms that are currently part of the standard-of-care, the electrocardiogram (ECG) and photo-plethysmogram (PPG), by adapting a deep learning architecture designed for image segmentation. Using over 150,000 min of data collected at two separate health systems from 463 patients, we demonstrate that our model provides a highly accurate prediction of the continuous ABP waveform (root mean square error 5.823 (95% CI 5.806–5.840) mmHg), as well as the derived systolic (mean difference 2.398 ± 5.623 mmHg) and diastolic blood pressure (mean difference − 2.497 ± 3.785 mmHg) compared to arterial line measurements. Our approach can potentially be used to measure blood pressure continuously and non-invasively for all patients in the acute care setting, without the need for any additional instrumentation beyond the current standard-of-care.

One of the core challenges in applying machine learning and artificial intelligence to medicine is the limited availability of annotated medical data. Unlike in other applications of machine learning, where an abundance of labeled data is available, the labeling and annotation of medical data and images require a major effort of manual work by expert clinicians who do not have the time to annotate manually. In this work, we propose a new deep learning technique (SLIVER-net), to predict clinical features from 3-dimensional volumes using a limited number of manually annotated examples. SLIVER-net is based on transfer learning, where we borrow information about the structure and parameters of the network from publicly available large datasets. Since public volume data are scarce, we use 2D images and account for the 3-dimensional structure using a novel deep learning method which tiles the volume scans, and then adds layers that leverage the 3D structure. In order to illustrate its utility, we apply SLIVER-net to predict risk factors for progression of age-related macular degeneration (AMD), a leading cause of blindness, from optical coherence tomography (OCT) volumes acquired from multiple sites. SLIVER-net successfully predicts these factors despite being trained with a relatively small number of annotated volumes (hundreds) and only dozens of positive training examples. Our empirical evaluation demonstrates that SLIVER-net significantly outperforms standard state-of-the-art deep learning techniques used for medical volumes, and its performance is generalizable as it was validated on an external testing set. In a direct comparison with a clinician panel, we find that SLIVER-net also outperforms junior specialists, and identifies AMD progression risk factors similarly to expert retina specialists.

What is the common denominator of consciousness across divergent regimes of cortical dynamics? Does consciousness show itself in decibels or in bits? To address these questions, we introduce a testbed for evaluating electroencephalogram (EEG) biomarkers of consciousness using dissociations between neural oscillations and consciousness caused by rare genetic disorders. Children with Angelman syndrome (AS) exhibit sleep-like neural dynamics during wakefulness. Conversely, children with duplication 15q11.2-13.1 syndrome (Dup15q) exhibit wake-like neural dynamics during non-rapid eye movement (NREM) sleep. To identify highly generalizable biomarkers of consciousness, we trained regularized logistic regression classifiers on EEG data from wakefulness and NREM sleep in children with AS using both entropy measures of neural complexity and spectral (i.e., neural oscillatory) EEG features. For each set of features, we then validated these classifiers using EEG from neurotypical (NT) children and abnormal EEGs from children with Dup15q. Our results show that the classification performance of entropy-based EEG biomarkers of conscious state is not upper-bounded by that of spectral EEG features, which are outperformed by entropy features. Entropy-based biomarkers of consciousness may thus be highly adaptable and should be investigated further in situations where spectral EEG features have shown limited success, such as detecting covert consciousness or anesthesia awareness.