Search results
Found 9788 matches for
Approximations to the Fisher Information Metric of Deep Generative Models for Out-Of-Distribution Detection
Likelihood-based deep generative models such as score-based diffusion models and variational autoencoders are state-of-the-art machine learning models approximating high-dimensional distributions of data such as images, text, or audio. One of many downstream tasks they can be naturally applied to is out-of-distribution (OOD) detection. However, seminal work by Nalisnick et al. which we reproduce showed that deep generative models consistently infer higher log-likelihoods for OOD data than data they were trained on, marking an open problem. In this work, we analyse using the gradient of a data point with respect to the parameters of the deep generative model for OOD detection, based on the simple intuition that OOD data should have larger gradient norms than training data. We formalise measuring the size of the gradient as approximating the Fisher information metric. We show that the Fisher information matrix (FIM) has large absolute diagonal values, motivating the use of chi-square distributed, layer-wise gradient norms as features. We combine these features to make a simple, model-agnostic and hyperparameter-free method for OOD detection which estimates the joint density of the layer-wise gradient norms for a given data point. We find that these layer-wise gradient norms are weakly correlated, rendering their combined usage informative, and prove that the layer-wise gradient norms satisfy the principle of (data representation) invariance. Our empirical results indicate that this method outperforms the Typicality test for most deep generative models and image dataset pairings.
From Big Data to the Clinic: Methodological and Statistical Enhancements to Implement the UK Biobank Imaging Framework in a Memory Clinic.
The analysis tools and statistical methods used in large neuroimaging research studies differ from those applied in clinical contexts, making it unclear whether these techniques can be translated to a memory clinic setting. The Oxford Brain Health Clinic (OBHC) was established in 2020 to bridge this gap between research studies and memory clinics. We optimised the UK Biobank imaging framework for the memory clinic setting by integrating enhanced quality control (QC) processes (MRIQC, QUAD, and DSE decomposition) and supplementary dementia-informed analyses (lobar volumes, NBM volumes, WMH classification, PSMD, cortical diffusion MRI metrics, and tract volumes) into the analysis pipeline. We explored associations between resultant imaging-derived phenotypes (IDPs) and clinical phenotypes in the OBHC patient population (N = 213), applying hierarchical FDR correction to account for multiple testing. 14%-24% of scans were flagged by automated QC tools, but upon visual inspection, only 0%-2.4% of outputs were excluded. The pipeline successfully generated 5683 IDPs aligned with UK Biobank and 110 IDPs targeted towards dementia-related changes. We replicated established associations and found novel associations between brain metrics and age, cognition, and dementia-related diagnoses. The imaging protocol is feasible, acceptable, and yields high-quality data that is usable for both clinical and research purposes. We validated the use of this methodology in a real-world memory clinic population, which demonstrates the potential of this enhanced pipeline to bridge the gap between big data studies and clinical settings.
Long-term outcomes of bilateral salpingo-oophorectomy in women with personal history of breast cancer.
OBJECTIVES: To investigate the association between bilateral salpingo-oophorectomy (BSO) and long-term health outcomes in women with a personal history of breast cancer. METHODS AND ANALYSIS: We used data on women diagnosed with invasive breast cancer between 1995 and 2019 from the National Cancer Registration Dataset (NCRD) in England. The data were linked to the Hospital Episode Statistics-Admitted Patient Care dataset to identify BSO delivery. Long-term health outcomes were selected from both datasets. Multivariable Cox regression was used to examine the associations, with BSO modelled as a time-dependent covariate. The associations were investigated separately by age at BSO. RESULTS: We identified 568 883 women, 23 401 of whom had BSO after the breast cancer diagnosis. There was an increased risk of total cardiovascular diseases with an HR of 1.10 (95% CI 1.04 to 1.16) in women who had BSO<55 years and 1.07 (95% CI 1.01 to 1.13) for women who had BSO≥55 years. There was an increased risk of ischaemic heart diseases, but there was no association with cerebrovascular diseases. BSO at any age was associated with an increased risk of depression (HR 1.20, 95% CI 1.12 to 1.28) and increased risk of second non-breast cancer in older women (HR 1.21, 95%CI 1.08 to 1.35). BSO in older women was associated with reduced risk of all-cause mortality (HR 0.92, 95% CI 0.87 to 096), but not in women who had BSO<55 years. CONCLUSION: In women with a personal history of breast cancer, BSO before and after the age of 55 years is associated with an increased risk of long-term outcomes. BSO after 55 years is associated with reduced all-cause mortality. Family history or genetic predisposition may confound these associations.
Convalescent plasma in patients admitted to hospital with COVID-19 (RECOVERY): a randomised, controlled, open-label, platform trial
ABSTRACT Background Treatment of COVID-19 patients with plasma containing anti-SARS-CoV-2 antibodies may have a beneficial effect on clinical outcomes. We aimed to evaluate the safety and efficacy of convalescent plasma in patients admitted to hospital with COVID-19. Methods In this randomised, controlled, open-label, platform trial (Randomised Evaluation of COVID-19 Therapy [RECOVERY]) several possible treatments are being compared with usual care in patients hospitalised with COVID-19 in the UK. Eligible and consenting patients were randomly allocated to receive either usual care plus high titre convalescent plasma or usual care alone. The primary outcome was 28-day mortality. Findings Between 28 May 2020 and 15 January 2021, 5795 patients were randomly allocated to receive convalescent plasma and 5763 to usual care alone. There was no significant difference in 28-day mortality between the two groups: 1398 (24%) of 5795 patients allocated convalescent plasma and 1408 (24%) of 5763 patients allocated usual care died within 28 days (rate ratio [RR] 1·00; 95% confidence interval [CI] 0·93 to 1·07; p=0·93). The 28-day mortality rate ratio was similar in all prespecified subgroups of patients, including in those patients without detectable SARS-CoV-2 antibodies at randomisation. Allocation to convalescent plasma had no significant effect on the proportion of patients discharged from hospital within 28 days (66% vs . 67%; rate ratio 0·98; 95% CI 0·94-1·03, p=0·50). Among those not on invasive mechanical ventilation at baseline, there was no significant difference in the proportion meeting the composite endpoint of progression to invasive mechanical ventilation or death (28% vs . 29%; rate ratio 0·99; 95% CI 0·93-1·05, p=0·79). Interpretation Among patients hospitalised with COVID-19, high-titre convalescent plasma did not improve survival or other prespecified clinical outcomes. Funding UK Research and Innovation (Medical Research Council) and National Institute of Health Research (Grant refs: MC_PC_19056; COV19-RECPLA).
Advancements in Fetal Heart Rate Monitoring: A Report on Opportunities and Strategic Initiatives for Better Intrapartum Care.
Cardiotocography (CTG), introduced in the 1960s, was initially expected to prevent hypoxia-related deaths and neurological injuries. However, more than five decades later, evidence supporting the evidence of intrapartum CTG in preventing neonatal and long-term childhood morbidity and mortality remains inconclusive. At the same time, shortcomings in CTG interpretation have been recognised as important contributory factors to rising caesarean section rates and missed opportunities for timely interventions. An important limitation is its high false-positive rate and poor specificity, which undermines reliably identifying foetuses at risk of hypoxia-related injuries. These shortcomings are compounded by the technology's significant intra- and interobserver variability, as well as the subjective and complex nature of fetal heart rate interpretation. However, human factors and other environmental factors are equally significant. Advancements in fetal heart rate monitoring are crucial to support clinicians in improving health outcomes for newborns and their mothers, while at the same time avoiding unnecessary operative deliveries. These limitations highlight the clinical need to enhance neonatal outcomes while minimising unnecessary interventions, such as instrumental deliveries or caesarean sections. We believe that achieving this requires a paradigm shift from subjective interpretation of complex and nonspecific fetal heart rate patterns to evidence-based, quantifiable solutions that integrate hardware, engineering and clinical perspectives. Such transformation necessitates an international, multidisciplinary effort encompassing the entire continuum of pregnancy care and the broader healthcare ecosystem, with emphasis on well-defined, actionable health outcomes. Achieving this will depend on collaborations between researchers, clinicians, medical device manufacturers and other relevant stakeholders. This expert review paper outlines the most relevant and promising directions for research and strategic initiatives to address current challenges in fetal heart rate monitoring. Key themes include advancements in computerised fetal heart rate monitoring, the application of big data and artificial intelligence, innovations in home and remote monitoring and consideration of human factors.
Gated Self Attention Convolutional Neural Networks for Predicting Adverse Birth Outcomes
Early detection of adverse birth outcomes is vital as they are major contributors to neonatal mortality and irreversible neurological complications in infants. These outcomes are typically linked to impaired blood and oxygen flow to the baby brain during or shortly after labour, making its early detection vital. Monitoring fetal heart rate (FHR) is crucial in identifying and capturing these complications. This study proposes a deep learning (DL) framework for enhancing the early detection of the babies at risk, leveraging both raw FHR signals and standard cardiotocography (CTG) features. Unlike traditional methods that primarily focus on abnormal CTG traces (but not birth outcomes), this approach, backed by a substantial cohort of patient records, demonstrates the potential of DL in predicting actual adverse outcomes as early as possible. The DL model combines a convolutional mechanism with a self-attention network, enhanced by a gating mechanism for more accurate feature learning. The investigated DL architecture is trained on a dataset of over 37,000 births, including 1,291 abnormal ones, and is evaluated on a holdout set of 6,459 births, as well as the open-access CTU-CHB CTG dataset of 552 births. The proposed DL model demonstrates superior diagnostic accuracy, outperforming state-of-the-art baseline methods and clinical benchmarks. It achieved sensitivity of 49.08% (95% CI, 46.01-53.36%) at 15% false positive rate (FPR), compared to the clinical benchmark sensitivity of 37.70% (33.10-42.30%) and a previous model's 32.60% (28.20-37.30%) at a similar FPR.