Search

We argue that artificial networks are explainable and offer a novel theory of interpretability. Two sets of conceptual questions are prominent in theoretical engagements with artificial neural networks, especially in the context of medical artificial intelligence: (1) Are networksexplainable, and if so, what does it mean to explain the output of a network? And (2) what does it mean for a network to beinterpretable? We argue that accounts of “explanation” tailored specifically to neural networks have ineffectively reinvented the wheel. In response to (1), we show how four familiar accounts of explanation apply to neural networks as they would to any scientific phenomenon. We diagnose the confusion about explaining neural networks within the machine learning literature as an equivocation on “explainability,” “understandability” and “interpretability.” To remedy this, we distinguish between these notions, and answer (2) by offering a theory and typology of interpretation in machine learning. Interpretation is something one does to an explanation with the aim of producing another, more understandable, explanation. As with explanation, there are various concepts and methods involved in interpretation:TotalorPartial,GlobalorLocal, andApproximativeorIsomorphic. Our account of “interpretability” is consistent with uses in the machine learning literature, in keeping with the philosophy of explanation and understanding, and pays special attention to medical artificial intelligence systems.

The challenge of survival prediction is ubiquitous in industry and medicine. Few methods are available for survival prediction of time varying data. Here we propose a novel method for this problem, using a random forest of survival trees for left truncated and right-censored data. We demonstrated the advantage of our method on prediction of breast cancer and prostate gland cancer risk among healthy individuals by analyzing routine laboratory measurements, vital signs and age. We analyzed electronic medical records of 20,317 healthy individuals who underwent routine checkups and identified those who later developed cancer. In cross-validation, our method predicted future prostate and breast cancers six months before diagnosis with an area under the ROC curve of 0.62±0.05 and 0.6±0.03 respectively, outperforming standard random forest, Cox-regression model and a single survival tree. Our results suggest that computational analysis of data on healthy individuals can improve the detection of those at risk of future cancer development.

Differential diagnosis between acute viral and bacterial infection is an emerging common challenge for a physician in the emergency department. Serum C-reactive protein (CRP) is used to support diagnosis of bacterial infection, but in patients admitted with low CRP, its ability to discriminate between viral and bacterial infections is limited. We aimed to use two consecutive CRP measurements in order to improve differential diagnosis between bacterial and viral infection. A single-center retrospective cohort (n = 1629) study of adult patients admitted to the emergency department with a subsequent microbiological confirmation of either viral or bacterial infection. Trend of CRP was defined as the absolute difference between the first two measurements of CRP divided by the time between them, and we investigated the ability of this parameter to differentiate between viral and bacterial infection. In patients with relatively low initial CRP concentration (

Transformative technologies are enabling the construction of three-dimensional maps of tissues with unprecedented spatial and molecular resolution. Over the next seven years, the NIH Common Fund Human Biomolecular Atlas Program (HuBMAP) intends to develop a widely accessible framework for comprehensively mapping the human body at single-cell resolution by supporting technology development, data acquisition, and detailed spatial mapping. HuBMAP will integrate its efforts with other funding agencies, programs, consortia, and the biomedical research community at large towards the shared vision of a comprehensive, accessible three-dimensional molecular and cellular atlas of the human body, in health and under various disease conditions., HuBMAP supports technology development, data acquisition, and spatial analyses to generate comprehensive molecular and cellular three-dimensional tissue maps.

Modern problems in statistics often include estimators of high computational complexity and with complicated distributions. Statistical inference on such estimators usually relies on asymptotic nor...

Background Imbalanced autonomic nervous system (ANS) activity is associated with poor cardiovascular outcome. However, clinically validated biomarkers to assess parasympathetic function are not yet available. We sought to evaluate parasympathetic dysfunction by measuring serum cholinesterase activity and to determine its relationship to high sensitive cardiac troponin T (hs-cTnT) as well as traditional non-invasive parameters of ANS function during exercise in apparently healthy individuals. Methods We enrolled 1526 individuals (mean age 49 ± 11 yr., 75% men) from the Tel Aviv Medical Center Inflammation Survey (TAMCIS). We used the acetylcholine (ACh) analog acetylthiocholine (ATCh) as a substrate that is hydrolyzed by both ACh degrading enzymes and reflects the total serum capacity for acetylcholine hydrolysis, referred to as cholinergic status (CS). All subjects performed a cardiac stress test reviewed on the spot by a cardiologist and multiple physiological and metabolic parameters including hs-cTnT were measured. Results CS values at rest predicted multiple exercise-hemodynamic changes. Heart rate recovery after exercise was inversely correlated to CS values (p 5 ng/L) presented with elevated CS levels compared to women with undetectable levels; 1423 ± 272.5 vs 1347 ± 297.9 (p = 0.02). An opposite trend was observed in men. Metabolic dysfunction parameters were also associated with CS elevation in both men and women. Conclusions Parasympathetic dysfunction assessed by total serum cholinesterase activity predicts hemodynamic changes during exercise. CS is also associated with hs-cTnT detection in women and inversely so in men. Future studies to assess the potential clinical use of this new sex-specific biomarker in cardiovascular disease risk stratification are warranted. Graphical Abstract Electronic supplementary material The online version of this article (10.1186/s10020-018-0063-0) contains supplementary material, which is available to authorized users.

Background Patient-facing digital health tools have been promoted to help patients manage concerns related to COVID-19 and to enable remote care and self-care during the COVID-19 pandemic. It has also been suggested that these tools can help further our understanding of the clinical characteristics of this new disease. However, there is limited information on the characteristics and use patterns of these tools in practice. Objective The aims of this study are to describe the characteristics of people who use digital health tools to address COVID-19–related concerns; explore their self-reported symptoms and characterize the association of these symptoms with COVID-19; and characterize the recommendations provided by digital health tools. Methods This study used data from three digital health tools on the K Health app: a protocol-based COVID-19 self-assessment, an artificial intelligence (AI)–driven symptom checker, and communication with remote physicians. Deidentified data were extracted on the demographic and clinical characteristics of adults seeking COVID-19–related health information between April 8 and June 20, 2020. Analyses included exploring features associated with COVID-19 positivity and features associated with the choice to communicate with a remote physician. Results During the period assessed, 71,619 individuals completed the COVID-19 self-assessment, 41,425 also used the AI-driven symptom checker, and 2523 consulted with remote physicians. Individuals who used the COVID-19 self-assessment were predominantly female (51,845/71,619, 72.4%), with a mean age of 34.5 years (SD 13.9). Testing for COVID-19 was reported by 2901 users, of whom 433 (14.9%) reported testing positive. Users who tested positive for COVID-19 were more likely to have reported loss of smell or taste (relative rate [RR] 6.66, 95% CI 5.53-7.94) and other established COVID-19 symptoms as well as ocular symptoms. Users communicating with a remote physician were more likely to have been recommended by the self-assessment to undergo immediate medical evaluation due to the presence of severe symptoms (RR 1.19, 95% CI 1.02-1.32). Most consultations with remote physicians (1940/2523, 76.9%) were resolved without need for referral to an in-person visit or to the emergency department. Conclusions Our results suggest that digital health tools can help support remote care and self-management of COVID-19 and that self-reported symptoms from digital interactions can extend our understanding of the symptoms associated with COVID-19.

BACKGROUND Patient-facing digital health tools have been promoted to help patients manage concerns related to COVID-19 and to enable remote care and self-care during the COVID-19 pandemic. It has also been suggested that these tools can help further our understanding of the clinical characteristics of this new disease. However, there is limited information on the characteristics and use patterns of these tools in practice. OBJECTIVE The aims of this study are to describe the characteristics of people who use digital health tools to address COVID-19–related concerns; explore their self-reported symptoms and characterize the association of these symptoms with COVID-19; and characterize the recommendations provided by digital health tools. METHODS This study used data from three digital health tools on the K Health app: a protocol-based COVID-19 self-assessment, an artificial intelligence (AI)–driven symptom checker, and communication with remote physicians. Deidentified data were extracted on the demographic and clinical characteristics of adults seeking COVID-19–related health information between April 8 and June 20, 2020. Analyses included exploring features associated with COVID-19 positivity and features associated with the choice to communicate with a remote physician. RESULTS During the period assessed, 71,619 individuals completed the COVID-19 self-assessment, 41,425 also used the AI-driven symptom checker, and 2523 consulted with remote physicians. Individuals who used the COVID-19 self-assessment were predominantly female (51,845/71,619, 72.4%), with a mean age of 34.5 years (SD 13.9). Testing for COVID-19 was reported by 2901 users, of whom 433 (14.9%) reported testing positive. Users who tested positive for COVID-19 were more likely to have reported loss of smell or taste (relative rate [RR] 6.66, 95% CI 5.53-7.94) and other established COVID-19 symptoms as well as ocular symptoms. Users communicating with a remote physician were more likely to have been recommended by the self-assessment to undergo immediate medical evaluation due to the presence of severe symptoms (RR 1.19, 95% CI 1.02-1.32). Most consultations with remote physicians (1940/2523, 76.9%) were resolved without need for referral to an in-person visit or to the emergency department. CONCLUSIONS Our results suggest that digital health tools can help support remote care and self-management of COVID-19 and that self-reported symptoms from digital interactions can extend our understanding of the symptoms associated with COVID-19.

Circulating tumor-derived DNA (ctDNA) can be used to monitor cancer dynamics noninvasively. Detection of ctDNA can be challenging in patients with low-volume or residual disease, where plasma contains very few tumor-derived DNA fragments. We show that sensitivity for ctDNA detection in plasma can be improved by analyzing hundreds to thousands of mutations that are first identified by tumor genotyping. We describe the INtegration of VAriant Reads (INVAR) pipeline, which combines custom error-suppression methods and signal-enrichment approaches based on biological features of ctDNA. With this approach, the detection limit in each sample can be estimated independently based on the number of informative reads sequenced across multiple patient-specific loci. We applied INVAR to custom hybrid-capture sequencing data from 176 plasma samples from 105 patients with melanoma, lung, renal, glioma, and breast cancer across both early and advanced disease. By integrating signal across a median of >105 informative reads, ctDNA was routinely quantified to 1 mutant molecule per 100,000, and in some cases with high tumor mutation burden and/or plasma input material, to parts per million. This resulted in median area under the curve (AUC) values of 0.98 in advanced cancers and 0.80 in early-stage and challenging settings for ctDNA detection. We generalized this method to whole-exome and whole-genome sequencing, showing that INVAR may be applied without requiring personalized sequencing panels so long as a tumor mutation list is available. As tumor sequencing becomes increasingly performed, such methods for personalized cancer monitoring may enhance the sensitivity of cancer liquid biopsies.

Metabolic syndrome (MetS) and high-sensitivity cardiac troponin T (hs-TnT) are associated with higher risk for cardiovascular diseases (CVD). Our aim was to assess the relation between hs-TnT elevation and MetS in a general population sample.Individuals participating in an annual health survey program between 2010 and 2016 were included in the study. Blood samples including hs-TnT levels were collected. The study population was divided into three groups based on hs-TnT levels - undetectable (5 ng/L), intermediate (5-14 ng/L) and elevated (14 ng/L).A total of 5994 subjects were included in the study, the mean age was 48.5 and 4336 (72%) were males. Compared with subjects with undetectable hs-TnT the prevalence of MetS was higher in those with detectable and elevated levels - 392 (10%) vs. 270 (15%) and 51 (33%), respectively (p 0.001). In a multivariate model adjusted for age, gender and multiple co-morbidities, the number of MetS components and presence of MetS were significantly associated with an increased risk for detectable hs-TnT levels (OR = 1.02 {for each component}; 95% CI [1.00-1.05], p = 0.04) and (OR = 1.13; 95% CI [1.07-1.2], p 0.001) respectively. Only the waist, glucose and hypertension components of the MetS were significantly associated with elevated troponin.The MetS and its distinct components have a cumulative impact on hs-TnT levels in apparently healthy subjects.

Estimation of heritability is an important task in genetics. The use of linear mixed models (LMMs) to determine narrow-sense single-nucleotide polymorphism (SNP)-heritability and related quantities has received much recent attention, due of its ability to account for variants with small effect sizes. Typically, heritability estimation under LMMs uses the restricted maximum likelihood (REML) approach. The common way to report the uncertainty in REML estimation uses standard errors (SEs), which rely on asymptotic properties. However, these assumptions are often violated because of the bounded parameter space, statistical dependencies, and limited sample size, leading to biased estimates and inflated or deflated confidence intervals (CIs). In addition, for larger data sets (e.g., tens of thousands of individuals), the construction of SEs itself may require considerable time, as it requires expensive matrix inversions and multiplications. Here, we present FIESTA (Fast confidence IntErvals using STochastic Approximation), a method for constructing accurate CIs. FIESTA is based on parametric bootstrap sampling, and, therefore, avoids unjustified assumptions on the distribution of the heritability estimator. FIESTA uses stochastic approximation techniques, which accelerate the construction of CIs by several orders of magnitude, compared with previous approaches as well as to the analytical approximation used by SEs. FIESTA builds accurate CIs rapidly, for example, requiring only several seconds for data sets of tens of thousands of individuals, making FIESTA a very fast solution to the problem of building accurate CIs for heritability for all data set sizes.

Circulating tumor-derived DNA (ctDNA) can be used to monitor cancer dynamics noninvasively. Patients with small tumors have few copies of ctDNA in plasma, resulting in limited sensitivity to detect low-volume or residual disease. We show that sampling limitations can be overcome and sensitivity for ctDNA detection can be improved by massively parallel sequencing when hundreds to thousands of mutations are identified by tumor genotyping. We describe the INtegration of VAriant Reads (INVAR) analysis pipeline, which combines patient-specific mutation lists with both custom error-suppression methods and signal enrichment based on biological features of ctDNA. In this framework, the sensitivity can be estimated independently for each sample based on the number of informative reads, which is the product of the number of mutations analyzed and the average depth of unique sequencing reads. We applied INVAR to deep sequencing data generated by custom hybrid-capture panels, and showed that when ~106 informative reads were obtained INVAR allowed detection of tumor-derived DNA fractions to parts per million (ppm). In serial samples from patients with advanced melanoma on treatment, we detected ctDNA when imaging confirmed tumor volume of ~1cm3. In patients with resected early-stage melanoma, ctDNA was detected in 40% of patients who later relapsed, with higher rates of detection when more informative reads were obtained. We further demonstrated that INVAR can be generalized and allows improved detection of ctDNA from whole-exome and low-depth whole-genome sequencing data.

Understanding the genetic and molecular drivers of phenotypic heterogeneity across individuals is central to biology. As new technologies enable fine-grained and spatially resolved molecular profiling, we need new computational approaches to integrate data from the same organ across different individuals into a consistent reference and to construct maps of molecular and cellular organization at histological and anatomical scales. Here, we review previous efforts and discuss challenges involved in establishing such a common coordinate framework, the underlying map of tissues and organs. We focus on strategies to handle anatomical variation across individuals and highlight the need for new technologies and analytical methods spanning multiple hierarchical scales of spatial resolution., National Institutes of Health (U.S.). The Human BioMolecular Atlas Program (Grant NIH 1OT2OD026673- 01), National Institutes of Health (U.S.). New Innovator Award (1DP2HG009623- 01), Chan Zuckerberg Initiative (Grant HCA2-A-1708-02755)

Background Despite advances in prevention strategies the diagnosis of neonatal sepsis and clinical decision making remains challenging. Empirical antibiotic treatment is given to neonates when sepsis is suspected. However, clinical and laboratory signs are generally unspecific and most neonates who receive antibiotics are not ultimately diagnosed with sepsis. Many physicians view empirical antibiotics as the safest course of action in cases of equivocal clinical presentation. The long term effects of early gut flora modification are poorly understood but some scientists suggested this may alter activation of genes involved in modulating immune responses. This study aims to provide some insight into the level of risk associated with typical indications for neonatal septic workup at our institution. Methods We conducted a retrospective case-control study. Infants born at ≥35 weeks gestation who received empirical antibiotics over a three months period were included along with all infants who were recorded to have culture positive sepsis in a ten year period. Three outcome groups were defined: (1) Culture positive sepsis (N=43) (2) Suspected culture negative sepsis (N=5) and (3) No sepsis (N=97). Rates of clinical symptoms and exposure to maternal and neonatal risk factors were compared. P-values were calculated using a test for equality of proportions implemented in the R programming language. Results There was a statistically significant increase in red flag clinical features (mechanical ventilation, seizures, respiratory distress starting more than 4 hours after delivery and signs of shock) in the culture positive sepsis group compared to the no sepsis group (p Conclusion Respiratory distress and suspected chorioamnionitis were the most common indications for a sepsis workup but neither was significantly more common in the culture positive sepsis group. Observation and repeated evaluation may be suitable for infants with equivocal presentation. Critically ill infants with red flag clinical features and infants with a greater number of clinical symptoms should have a blood culture taken and IV antibiotics commenced without delay.

Summary Accurate measurement of clonal genotypes, mutational processes, and replication states from individual tumor-cell genomes will facilitate improved understanding of tumor evolution. We have developed DLP+, a scalable single-cell whole-genome sequencing platform implemented using commodity instruments, image-based object recognition, and open source computational methods. Using DLP+, we have generated a resource of 51,926 single-cell genomes and matched cell images from diverse cell types including cell lines, xenografts, and diagnostic samples with limited material. From this resource we have defined variation in mitotic mis-segregation rates across tissue types and genotypes. Analysis of matched genomic and image measurements revealed correlations between cellular morphology and genome ploidy states. Aggregation of cells sharing copy number profiles allowed for calculation of single-nucleotide resolution clonal genotypes and inference of clonal phylogenies and avoided the limitations of bulk deconvolution. Finally, joint analysis over the above features defined clone-specific chromosomal aneuploidy in polyclonal populations., Graphical Abstract, Highlights • Scaled method and resource of > 50K single-cell whole genomes from diverse cell types • Clonal merging can resolve clone specific mutations to single-nucleotide level • Image analysis of single cells permits correlation of morphology and genome features • Clonal replication states and rare aneuploidy patterns of single cells measured, A high-throughput method for amplication-free single-cell whole-genome sequencing can be scaled up to analyze tens of thousands of cells from different tissues and clinical sample types and identifies replication states, aneuploidies, and subclonal mutations.

The metabolic syndrome (MetS) is associated with overweight and abdominal obesity. Our aim was to use longitudinal measurements to provide clinically relevant information on the relative influence of changes in body mass index (BMI), waist circumference (WC), and weekly physical exercise duration on the development of each of the MetS components. We analyzed data collected at the Tel-Aviv Medical Center Inflammation Survey (TAMCIS). Apparently healthy individuals with two consecutive visits that were not treated for any metabolic criteria were included in this study. We analyzed the influence of changes in BMI, WC, and time engaged in physical exercise on the change in each of the components of the metabolic syndrome using linear regressions. Included were 7532 individuals (5431 men, 2101 women) with 2 years follow-up. Participants who gained two BMI points, had the mean number of criteria increase from 1.07 to 1.52, while participants who lost two BMI points, decreased from 1.64 to 1.16. A long-term analysis over 5 years showed similar results. Furthermore, an increase in WC was independently associated with increased severity of each of the other components, when controlling for increase in BMI. Increase in weekly exercise duration had a small but statistically significant favorable effect on blood triglycerides and HDL levels, but not on blood pressure or HbA1C. Changes in BMI and WC are highly associative with the likelihood and severity of the MetS independently of the baseline levels, suggesting that obese individuals can substantially improve their MetS prognosis by losing both body weight and abdominal fat.

BACKGROUND The exercise ECG stress test (EST) is still the first step of work-up in intermediate risk patients in many clinical scenarios. High-sensitive cardiac troponin T (hs-cTnT) elevation is related to future cardiovascular events in the general population and in patients with ischaemic heart disease. The relation between these 2 tests is not well described. MATERIALS AND METHODS A total of 2780 participants from the Tel-Aviv Medical Center Inflammation Survey cohort (mean age 49 years, 79% men) were analysed. Multiple physiologic and metabolic parameters including hs-cTnT were collected. All participants completed an EST manually reviewed by a cardiologist. RESULTS A positive EST was documented in 224 subjects (8%). The majority (91%) of participants with hs-cTnT levels of 5-14 ng/L had a negative EST as well as 89.3% of subjects with levels >14 ng/L. The proportion of subjects with a positive EST and detectable hs-cTnT levels (>5 ng/L) was not significantly greater compared to those with a negative EST (53.1% vs 46.2%, respectively, P = .09). CONCLUSION Among subjects referred for EST as part of an annual health survey, we found no significant association between EST results to hs-cTnT detection.

Estimation of heritability is an important task in genetics. The use of linear mixed models (LMMs) to determine narrow-sense SNP-heritability and related quantities has received much recent attention, due of its ability to account for variants with small effect sizes. Typically, heritability estimation under LMMs uses the restricted maximum likelihood (REML) approach. The common way to report the uncertainty in REML estimation uses standard errors (SE), which rely on asymptotic properties. However, these assumptions are often violated because of the bounded parameter space, statistical dependencies, and limited sample size, leading to biased estimates and inflated or deflated confidence intervals. In addition, for larger datasets (e.g., tens of thousands of individuals), the construction of SEs itself may require considerable time, as it requires expensive matrix inversions and multiplications.