Matching science to reality: how to deploy a participant-driven digital brain health platform

Introduction

Advances in digital technologies for health research enable opportunities for digital phenotyping of individuals in research and clinical settings. Beyond providing opportunities for advanced data analytics with data science and machine learning approaches, digital technologies offer solutions to several of the existing barriers in research practice that have resulted in biased samples.

Methods

A participant-driven, precision brain health monitoring digital platform has been introduced to two longitudinal cohort studies, the Boston University Alzheimer's Disease Research Center (BU ADRC) and the Bogalusa Heart Study (BHS). The platform was developed with prioritization of digital data in native format, multiple OS, validity of derived metrics, feasibility and usability. A platform including nine remote technologies and three staff-guided digital assessments has been introduced in the BU ADRC population, including a multimodal smartphone application also introduced to the BHS population. Participants select which technologies they would like to use and can manipulate their personal platform and schedule over time.

Results

Participants from the BU ADRC are using an average of 5.9 technologies to date, providing strong evidence for the usability of numerous digital technologies in older adult populations. Broad phenotyping of both cohorts is ongoing, with the collection of data spanning cognitive testing, sleep, physical activity, speech, motor activity, cardiovascular health, mood, gait, balance, and more. Several challenges in digital phenotyping implementation in the BU ADRC and the BHS have arisen, and the protocol has been revised and optimized to minimize participant burden while sustaining participant contact and support.

Discussion

The importance of digital data in its native format, near real-time data access, passive participant engagement, and availability of technologies across OS has been supported by the pattern of participant technology use and adherence across cohorts. The precision brain health monitoring platform will be iteratively adjusted and improved over time. The pragmatic study design enables multimodal digital phenotyping of distinct clinically characterized cohorts in both rural and urban U.S. settings.

Abstract MP03: Adiponectin Levels in Young Adulthood and the Association With Midlife Cognition: The Bogalusa Heart Study

Objective: Adiponectin (ApN), has anti-inflammatory and anti-atherogenic properties. Low levels of ApN are found even after normalization of glucose in patients with diabetes, and are implicated in atherosclerosis, and dementia onset but results vary across the lifespan. We assessed the relation of ApN levels in young adulthood, with midlife atherosclerosis, measured by carotid intima media thickness (cIMT), and with cognitive function among Black and White individuals from the Bogalusa Heart Study (BHS).

Methods: A total of 604 participants from the BHS, a community cohort in rural Louisiana, were examined for ApN levels with lowest quartile of ApN levels (<5.4 μg/mL) defined as low-ApN. Structural equation models assessed mediation via the temporal association of low-ApN (age 36±4) with midlife (age 49±5) cIMT, dichotomized above the 50 th percentile (>0.88mm), and a global cognitive score(GCS) standardized for age, race and sex. Adjusted models included education attendance.

Results: Individuals with low-ApN were more likely men [ 56.2% (77), p<0.001] and White [54.7% (75), p<0.001]. No differences in education were found. There was a direct association between low-ApN and GCS (P=0.025), low-ApN and cIMT ≥ 50 th (P=0.001), and cIMT ≥ 50 th and GCS (P=0.002). ( Figure1) In mediation analysis, the ratio of Indirect effect/Total effect was 0.108, indicating that ~11% of the effect of low-ApN on GCS was mediated by cIMT. Education was not associated with low-ApN.

Conclusion: Low-ApN levels in early adulthood was associated with poor cognition after ~10 years of follow-up, and the association was partially mediated by subclinical atherosclerosis. This suggests that ApN may play a role in neurodegeneration mechanisms and contribute to the development of dementia, underscoring the importance of addressing cardiometabolic influences earlier in the lifespan to prevent/delay cognitive decline.

Alleviating iatrogenic effects of paclitaxel via anti-inflammatory treatment

Background

Paclitaxel is touted as an essential medicine due to its extensive use as a chemotherapeutic for various cancers and an antiproliferative agent for restenosis. Due to recent concerns related to long-term mortality, paclitaxel (PTX)-based endovascular therapy is now surrounded by controversies.

Objective

Examine the inflammatory mediators driven by the systemic administration of PTX and explore the means to suppress these effects.

Methods

RNAseq analysis, cell and mouse models.

Results

RNAseq analysis of primary human endothelial cells (ECs) treated with PTX demonstrated transcriptional perturbations of a set of pro-inflammatory mediators, including monocyte chemoattractant protein-1 (MCP-1) and CD137, which were validated in EC lysates. These perturbations were abrogated with dexamethasone, a prototypic anti-inflammatory compound. The media of ECs pre-treated with PTX showed a significant increase in MCP-1 levels, which were reverted to baseline levels with DEX treatment. A group of mice harvested at different time points after PTX injection were analyzed for immediate and delayed effects of PTX. A 3-fold increase in MCP-1 was noted in blood and aortic ECs after 12 hours of PTX treatment. Similar changes in CD137 and downstream mediators such as tissue factor, VCAM-1 and E-selectin were noted in aortic ECs.

Conclusions

Our study shows that systemic PTX exposure upregulates atherothrombotic markers, and co-delivery of DEX can subdue the untoward toxic effects. Long-term studies are needed to probe the mechanisms driving systemic complications of PTX-based therapies and evaluate the clinical potential of DEX to mitigate risk.

Understudied factors in drug-coated balloon design and evaluation: A biophysical perspective

Drug-coated balloon (DCB) percutaneous interventional therapy allows for durable reopening of the narrowed lumen via physical tissue expansion and local anti-restenosis drug delivery, providing an alternative to traditional uncoated balloons or a permanent indwelling implant such as a conventional metallic drug-eluting stent. While DCB-based treatment of peripheral arterial disease (PAD) has been incorporated into clinical guidelines, DCB use has been recently curtailed due to reports that showed evidence of increased mortality risk in patients treated with paclitaxel (PTX)-coated balloons. Given the United States Food and Drug Administration's 2019 consequent warning regarding PTX-eluting DCBs and the subsequent marked reduction in clinical DCB use, there is now a critical need to better understand the compositional and mechanical factors underlying DCB efficacy and safety. Most work to date on DCB refinement has focused on designing both the enabling balloon catheter and alternate coatings composed of various drugs and excipients, followed by device evaluation in preclinical and clinical studies. We contend that improvement in DCB performance will require a better understanding of the biophysical factors operative during and following balloon deployment, and moreover that the elaboration and demonstrated control of these factors are needed to address current concerns with DCB use. This article provides a perspective on the biophysical interactions that govern DCB performance and offers new design strategies for the development of next-generation DCB devices.

Midlife Neuropsychological Profiles and Associated Vascular Risk: The Bogalusa Heart Study

BACKGROUND

Individuals with Alzheimer's disease (AD) often present with coexisting vascular pathology that is expressed to different degrees and can lead to clinical heterogeneity.

OBJECTIVE

To examine the utility of unsupervised statistical clustering approaches in identifying neuropsychological (NP) test performance subtypes that closely correlate with carotid intima-media thickness (cIMT) in midlife.

METHODS

A hierarchical agglomerative and k-means clustering analysis based on NP scores (standardized for age, sex, and race) was conducted among 1,203 participants (age 48±5.3 years) from the Bogalusa Heart Study. Regression models assessed the association between cIMT ≥50th percentile and NP profiles, and global cognitive score (GCS) tertiles for sensitivity analysis.

RESULTS

Three NP profiles were identified: Mixed-low performance [16%, n = 192], scores ≥1 SD below the mean on immediate, delayed free recall, recognition verbal memory, and information processing; Average [59%, n = 704]; and Optimal [26%, n = 307] NP performance. Participants with greater cIMT were more likely to have a Mixed-low profile [OR = 3.10, 95% CI (2.13, 4.53), p < 0.001] compared to Optimal. After adjusting for education and cardiovascular (CV) risks, results remained. The association with GCS tertiles was more attenuated [lowest (34%, n = 407) versus highest (33%, n = 403) tertile: adjusted OR = 1.66, 95% CI (1.07, 2.60), p = 0.024].

CONCLUSION

As early as midlife, individuals with higher subclinical atherosclerosis were more likely to be in the Mixed-low profile, underscoring the potential malignancy of CV risk as related to NP test performance, suggesting that classification approaches may aid in identifying those at risk for AD/vascular dementia spectrum illness.

Fusion of Low-Level Descriptors of Digital Voice Recordings for Dementia Assessment

Digital voice recordings can offer affordable, accessible ways to evaluate behavior and function. We assessed how combining different low-level voice descriptors can evaluate cognitive status. Using voice recordings from neuropsychological exams at the Framingham Heart Study, we developed a machine learning framework fusing spectral, prosodic, and sound quality measures early in the training cycle. The model's area under the receiver operating characteristic curve was 0.832 (±0.034) in differentiating persons with dementia from those who had normal cognition. This offers a data-driven framework for analyzing minimally processed voice recordings for cognitive assessment, highlighting the value of digital technologies in disease detection and intervention.

Relation of gait measures with mild unilateral knee pain during walking using machine learning

Gait alterations in those with mild unilateral knee pain during walking may provide clues to modifiable alterations that affect progression of knee pain and osteoarthritis (OA). To examine this, we applied machine learning (ML) approaches to gait data from wearable sensors in a large observational knee OA cohort, the Multicenter Osteoarthritis (MOST) study. Participants completed a 20-m walk test wearing sensors on their trunk and ankles. Parameters describing spatiotemporal features of gait and symmetry, variability and complexity were extracted. We used an ensemble ML technique ("super learning") to identify gait variables in our cross-sectional data associated with the presence/absence of unilateral knee pain. We then used logistic regression to determine the association of selected gait variables with odds of mild knee pain. Of 2066 participants (mean age 63.6 [SD: 10.4] years, 56% female), 21.3% had mild unilateral pain while walking. Gait parameters selected in the ML process as influential included step regularity, sample entropy, gait speed, and amplitude dominant frequency, among others. In adjusted cross-sectional analyses, lower levels of step regularity (i.e., greater gait variability) and lower sample entropy(i.e., lower gait complexity) were associated with increased likelihood of unilateral mild pain while walking [aOR 0.80 (0.64-1.00) and aOR 0.79 (0.66-0.95), respectively].

A Graph-Transformer for Whole Slide Image Classification

Deep learning is a powerful tool for whole slide image (WSI) analysis. Typically, when performing supervised deep learning, a WSI is divided into small patches, trained and the outcomes are aggregated to estimate disease grade. However, patch-based methods introduce label noise during training by assuming that each patch is independent with the same label as the WSI and neglect overall WSI-level information that is significant in disease grading. Here we present a Graph-Transformer (GT) that fuses a graph-based representation of an WSI and a vision transformer for processing pathology images, called GTP, to predict disease grade. We selected 4,818 WSIs from the Clinical Proteomic Tumor Analysis Consortium (CPTAC), the National Lung Screening Trial (NLST), and The Cancer Genome Atlas (TCGA), and used GTP to distinguish adenocarcinoma (LUAD) and squamous cell carcinoma (LSCC) from adjacent non-cancerous tissue (normal). First, using NLST data, we developed a contrastive learning framework to generate a feature extractor. This allowed us to compute feature vectors of individual WSI patches, which were used to represent the nodes of the graph followed by construction of the GTP framework. Our model trained on the CPTAC data achieved consistently high performance on three-label classification (normal versus LUAD versus LSCC: mean accuracy = 91.2 ± 2.5%) based on five-fold cross-validation, and mean accuracy = 82.3 ± 1.0% on external test data (TCGA). We also introduced a graph-based saliency mapping technique, called GraphCAM, that can identify regions that are highly associated with the class label. Our findings demonstrate GTP as an interpretable and effective deep learning framework for WSI-level classification.

Automated detection of mild cognitive impairment and dementia from voice recordings: A natural language processing approach

INTRODUCTION

Automated computational assessment of neuropsychological tests would enable widespread, cost-effective screening for dementia.

METHODS

A novel natural language processing approach is developed and validated to identify different stages of dementia based on automated transcription of digital voice recordings of subjects' neuropsychological tests conducted by the Framingham Heart Study (n = 1084). Transcribed sentences from the test were encoded into quantitative data and several models were trained and tested using these data and the participants' demographic characteristics.

RESULTS

Average area under the curve (AUC) on the held-out test data reached 92.6%, 88.0%, and 74.4% for differentiating Normal cognition from Dementia, Normal or Mild Cognitive Impairment (MCI) from Dementia, and Normal from MCI, respectively.

DISCUSSION

The proposed approach offers a fully automated identification of MCI and dementia based on a recorded neuropsychological test, providing an opportunity to develop a remote screening tool that could be adapted easily to any language.

Local Supply Of Postdischarge Care Options Tied To Hospital Readmission Rates

The extent to which patients' risk for readmission after a hospitalization is influenced by local availability of postdischarge care options is not currently known. We used national, hospital-level data to assess whether the supply of postdischarge care options in hospitals' catchment areas was associated with readmission rates for Medicare patients after hospitalizations for acute myocardial infarction, heart failure, or pneumonia. Overall, readmission rates were negatively associated with per capita supply of primary care physicians (-0.16 percentage points per standard deviation) and licensed nursing home beds (-0.09 percentage points per standard deviation). In contrast, readmission rates were positively associated with per capita supply of nurse practitioners (0.09 percentage points per standard deviation). Our results suggest potential modifications to the Hospital Readmissions Reduction Program to account for local health system characteristics when assigning penalties to hospitals.

Multimodal deep learning for Alzheimer's disease dementia assessment

Worldwide, there are nearly 10 million new cases of dementia annually, of which Alzheimer's disease (AD) is the most common. New measures are needed to improve the diagnosis of individuals with cognitive impairment due to various etiologies. Here, we report a deep learning framework that accomplishes multiple diagnostic steps in successive fashion to identify persons with normal cognition (NC), mild cognitive impairment (MCI), AD, and non-AD dementias (nADD). We demonstrate a range of models capable of accepting flexible combinations of routinely collected clinical information, including demographics, medical history, neuropsychological testing, neuroimaging, and functional assessments. We then show that these frameworks compare favorably with the diagnostic accuracy of practicing neurologists and neuroradiologists. Lastly, we apply interpretability methods in computer vision to show that disease-specific patterns detected by our models track distinct patterns of degenerative changes throughout the brain and correspond closely with the presence of neuropathological lesions on autopsy. Our work demonstrates methodologies for validating computational predictions with established standards of medical diagnosis.

Smartphone-Based Neuropsychological Assessment in Parkinson's Disease: Feasibility, Validity, and Contextually Driven Variability in Cognition

OBJECTIVES

The prevalence of neurodegenerative disorders demands methods of accessible assessment that reliably captures cognition in daily life contexts. We investigated the feasibility of smartphone cognitive assessment in people with Parkinson's disease (PD), who may have cognitive impairment in addition to motor-related problems that limit attending in-person clinics. We examined how daily-life factors predicted smartphone cognitive performance and examined the convergent validity of smartphone assessment with traditional neuropsychological tests.

METHODS

Twenty-seven nondemented individuals with mild-moderate PD attended one in-lab session and responded to smartphone notifications over 10 days. The smartphone app queried participants 5x/day about their location, mood, alertness, exercise, and medication state and administered mobile games of working memory and executive function.

RESULTS

Response rate to prompts was high, demonstrating feasibility of the approach. Between-subject reliability was high on both cognitive games. Within-subject variability was higher for working memory than executive function. Strong convergent validity was seen between traditional tests and smartphone working memory but not executive function, reflecting the latter's ceiling effects. Participants performed better on mobile working memory tasks when at home and after recent exercise. Less self-reported daytime sleepiness and lower PD symptom burden predicted a stronger association between later time of day and higher smartphone test performance.

CONCLUSIONS

These findings support feasibility and validity of repeat smartphone assessments of cognition and provide preliminary evidence of the effects of context on cognitive variability in PD. Further development of this accessible assessment method could increase sensitivity and specificity regarding daily cognitive dysfunction for PD and other clinical populations.

Carotid Intima-media Thickness and Midlife Cognitive Function: Impact of Race and Social Disparities in the Bogalusa Heart Study

BACKGROUND AND OBJECTIVE

Carotid intima-media thickness (c-IMT) is a measurement of atherosclerosis, a progressive disease that develops as early as childhood and has been linked with cognitive impairment and dementia in the elderly. However, the relationship between c-IMT and midlife cognitive function as well as race and social disparities on this relationship remains unclear. We examined the association between c-IMT and cognitive function in midlife among Black and White participants from a semi-rural community-based cohort in Bogalusa, Louisiana.

METHODS

In this cross-sectional analysis of participants from the Bogalusa Heart Study, linear regression models were used to determine the association between c-IMT dichotomized above the 50^th percentile (> 0.87mm), an a demographically standardized global cognitive (GCS) and individual cognitive domain-based z-scores. Stratified analyses were performed to evaluate the impact of race and the individual's education status.

RESULTS

A total of 1,217 participants (age 48 ± 5.28 years) were included, 66% (804) self-identified as White and 34% (413) as Black. Of those, 58% (708) were women, and 42% (509) were men Having a c-IMT > 50^th percentile was inversely associated with GCS ([B ± SE]: -0.39 ± 0.18, P=0.03), independent of cardiovascular risk factors (CVRFs) and achieved education. The effect remained significant in Black and White participants after adjustment for CVRFs (Blacks: [B ± SE]: -1.25 ± 0.45, P=0.005; Whites: [B ± SE]: -0.92 ± 0.35, P=0.008), but not for education. The interaction between c-IMT >50^th percentile and education was significant (P=0.03), and stratified analysis showed an association with GCS among those with lower achieved education ([B ± SE]: -0.81 ± 0.33, P=0.013) independent of major CVRFs.

DISCUSSION

Subclinical atherosclerosis, measured as c-IMT, was associated with worse midlife cognitive function, independent of major CVRFs. The association was buffered by education and may be stronger among Black than White participants, likely due to corresponding structural and social determinants. These findings underscore the importance of establishing preventive measures in midlife and suggest subclinical atherosclerosis as a potential target to prevent cognitive decline.

Deep learning for subtyping the Alzheimer's disease spectrum

In a recent article from Cell Reports Medicine, Kwak et al. generate novel insights about subtyping cognitively impaired individuals based on structural imaging. Quantifying heterogeneity in Alzheimer's disease via subtyping could help us harness new disease-modifying therapies and improve patient care by providing a more targeted approach.

Redefining and Validating Digital Biomarkers as Fluid, Dynamic Multi-Dimensional Digital Signal Patterns

"Digital biomarker" is a term broadly and indiscriminately applied and often limited in its conceptualization to mimic well-established biomarkers as defined and approved by regulatory agencies such as the United States Food and Drug Administration (FDA). There is a practical urgency to revisit the definition of a digital biomarker and expand it beyond current methods of identification and validation. Restricting the promise of digital technologies within the realm of currently defined biomarkers creates a missed opportunity. A whole new field of prognostic and early diagnostic digital biomarkers driven by data science and artificial intelligence can break the current cycle of high healthcare costs and low health quality that is being driven by today's chronic disease detection and treatment approaches. This new class of digital biomarkers will be dynamic and require developing new FDA approval pathways and next-generation gold standards.

Machine Learning Applications in Nephrology: A Bibliometric Analysis Comparing Kidney Studies to Other Medicine Subspecialities

RATIONALE & OBJECTIVES

Artificial intelligence driven by machine learning algorithms is being increasingly employed for early detection, disease diagnosis, and clinical management. We explored the use of machine learning-driven advancements in kidney research compared with other organ-specific fields.

STUDY DESIGN

Cross-sectional bibliometric analysis.

SETTING & PARTICIPANTS

ISI Web of Science database was queried using specific Medical Subject Headings (MeSH) terms about the organ system, journal International Standard Serial Number, and research methodology. In parallel, we screened the National Institutes of Health (NIH) RePORTER website to explore funded grants that proposed the use of machine learning as a methodology.

PREDICTORS

Number of publications using machine learning as a research method.

OUTCOME

Articles were characterized by research methodology among 5 organ systems (brain, heart, kidney, liver, and lung). Grants funded by NIH for machine learning were characterized by study sections.

ANALYTICAL APPROACH

Percentages of articles using machine learning and other research methodologies were compared among 5 organ systems.

RESULTS

Machine learning-based articles that are focused on the kidney accounted for 3.2% of the total relevant articles from the 5 organ systems. Specifically, brain research published over 19-fold higher number of articles than kidney research. As compared with machine learning, conventional statistical approaches such as the Cox proportional hazard model were used 9-fold higher in articles related to kidney research. In general, a lower utilization of machine learning-based approaches was observed in organ-specific specialty journals than the broad interdisciplinary journals. The digestive disease, kidney, and urology study sections funded 122 applications proposing machine learning-based approaches compared to 265 applications from the neurology, neuropsychology, and neuropathology study sections.

LIMITATIONS

Observational study.

CONCLUSIONS

Our analysis suggests lowest use of machine learning as a research tool among kidney researchers compared with other organ-specific researchers, underscoring a need to better inform the kidney research community about this emerging data analytic tool.

Detection of dementia on voice recordings using deep learning: a Framingham Heart Study

BACKGROUND

Identification of reliable, affordable, and easy-to-use strategies for detection of dementia is sorely needed. Digital technologies, such as individual voice recordings, offer an attractive modality to assess cognition but methods that could automatically analyze such data are not readily available.

METHODS AND FINDINGS

We used 1264 voice recordings of neuropsychological examinations administered to participants from the Framingham Heart Study (FHS), a community-based longitudinal observational study. The recordings were 73 min in duration, on average, and contained at least two speakers (participant and examiner). Of the total voice recordings, 483 were of participants with normal cognition (NC), 451 recordings were of participants with mild cognitive impairment (MCI), and 330 were of participants with dementia (DE). We developed two deep learning models (a two-level long short-term memory (LSTM) network and a convolutional neural network (CNN)), which used the audio recordings to classify if the recording included a participant with only NC or only DE and to differentiate between recordings corresponding to those that had DE from those who did not have DE (i.e., NDE (NC+MCI)). Based on 5-fold cross-validation, the LSTM model achieved a mean (±std) area under the receiver operating characteristic curve (AUC) of 0.740 ± 0.017, mean balanced accuracy of 0.647 ± 0.027, and mean weighted F1 score of 0.596 ± 0.047 in classifying cases with DE from those with NC. The CNN model achieved a mean AUC of 0.805 ± 0.027, mean balanced accuracy of 0.743 ± 0.015, and mean weighted F1 score of 0.742 ± 0.033 in classifying cases with DE from those with NC. For the task related to the classification of participants with DE from NDE, the LSTM model achieved a mean AUC of 0.734 ± 0.014, mean balanced accuracy of 0.675 ± 0.013, and mean weighted F1 score of 0.671 ± 0.015. The CNN model achieved a mean AUC of 0.746 ± 0.021, mean balanced accuracy of 0.652 ± 0.020, and mean weighted F1 score of 0.635 ± 0.031 in classifying cases with DE from those who were NDE.

CONCLUSION

This proof-of-concept study demonstrates that automated deep learning-driven processing of audio recordings of neuropsychological testing performed on individuals recruited within a community cohort setting can facilitate dementia screening.

Detection of dementia on voice recordings using deep learning: a Framingham Heart Study

BACKGROUND

Identification of reliable, affordable, and easy-to-use strategies for detection of dementia is sorely needed. Digital technologies, such as individual voice recordings, offer an attractive modality to assess cognition but methods that could automatically analyze such data are not readily available.

METHODS AND FINDINGS

We used 1264 voice recordings of neuropsychological examinations administered to participants from the Framingham Heart Study (FHS), a community-based longitudinal observational study. The recordings were 73 min in duration, on average, and contained at least two speakers (participant and examiner). Of the total voice recordings, 483 were of participants with normal cognition (NC), 451 recordings were of participants with mild cognitive impairment (MCI), and 330 were of participants with dementia (DE). We developed two deep learning models (a two-level long short-term memory (LSTM) network and a convolutional neural network (CNN)), which used the audio recordings to classify if the recording included a participant with only NC or only DE and to differentiate between recordings corresponding to those that had DE from those who did not have DE (i.e., NDE (NC+MCI)). Based on 5-fold cross-validation, the LSTM model achieved a mean (±std) area under the receiver operating characteristic curve (AUC) of 0.740 ± 0.017, mean balanced accuracy of 0.647 ± 0.027, and mean weighted F1 score of 0.596 ± 0.047 in classifying cases with DE from those with NC. The CNN model achieved a mean AUC of 0.805 ± 0.027, mean balanced accuracy of 0.743 ± 0.015, and mean weighted F1 score of 0.742 ± 0.033 in classifying cases with DE from those with NC. For the task related to the classification of participants with DE from NDE, the LSTM model achieved a mean AUC of 0.734 ± 0.014, mean balanced accuracy of 0.675 ± 0.013, and mean weighted F1 score of 0.671 ± 0.015. The CNN model achieved a mean AUC of 0.746 ± 0.021, mean balanced accuracy of 0.652 ± 0.020, and mean weighted F1 score of 0.635 ± 0.031 in classifying cases with DE from those who were NDE.

CONCLUSION

This proof-of-concept study demonstrates that automated deep learning-driven processing of audio recordings of neuropsychological testing performed on individuals recruited within a community cohort setting can facilitate dementia screening.

Deep-Learning-Driven Quantification of Interstitial Fibrosis in Digitized Kidney Biopsies

Interstitial fibrosis and tubular atrophy (IFTA) on a renal biopsy are strong indicators of disease chronicity and prognosis. Techniques that are typically used for IFTA grading remain manual, leading to variability among pathologists. Accurate IFTA estimation using computational techniques can reduce this variability and provide quantitative assessment. Using trichrome-stained whole-slide images (WSIs) processed from human renal biopsies, we developed a deep-learning framework that captured finer pathologic structures at high resolution and overall context at the WSI level to predict IFTA grade. WSIs (n = 67) were obtained from The Ohio State University Wexner Medical Center. Five nephropathologists independently reviewed them and provided fibrosis scores that were converted to IFTA grades: ≤10% (none or minimal), 11% to 25% (mild), 26% to 50% (moderate), and >50% (severe). The model was developed by associating the WSIs with the IFTA grade determined by majority voting (reference estimate). Model performance was evaluated on WSIs (n = 28) obtained from the Kidney Precision Medicine Project. There was good agreement on the IFTA grading between the pathologists and the reference estimate (κ = 0.622 ± 0.071). The accuracy of the deep-learning model was 71.8% ± 5.3% on The Ohio State University Wexner Medical Center and 65.0% ± 4.2% on Kidney Precision Medicine Project data sets. Our approach to analyzing microscopic- and WSI-level changes in renal biopsies attempts to mimic the pathologist and provides a regional and contextual estimation of IFTA. Such methods can assist clinicopathologic diagnosis.

Medical diagnosis at the point-of-care by portable high-field asymmetric waveform ion mobility spectrometry: a systematic review and meta-analysis

Non-invasive medical diagnosis by analysing volatile organic compounds (VOCs) at the point-of-care is becoming feasible owing to recent advances in portable instrumentation. A number of studies have assessed the performance of a state-of-the-art VOC analyser (micro-chip high-field asymmetric waveform ion mobility spectrometry, FAIMS) for medical diagnosis. However, a comprehensive meta-analysis is needed to investigate the overall diagnostic performance of these novel methods across different medical conditions. An electronic search was performed using the CAplus and MEDLINE database through the SciFinder platform. The review identified a total of 23 studies and 2312 individuals. Eighteen studies were used for meta-analysis. A pooled analysis found an overall sensitivity of 80% (95% CI, 74%-85%,I2= 62%), and specificity of 78% (95% CI, 70%-84%,I2= 80%), which corresponds to the overall diagnostic performance of micro-chip FAIMS across many different medical conditions. The diagnostic accuracy was particularly high for coeliac and inflammatory bowel disease (sensitivity and specificity from 74% to 97%). The overall diagnostic performance was similar across breath, urine, and faecal matrices with sparse logistic regression and random forests algorithms resulting in higher diagnostic accuracy. Sources of variability likely arise from differences in sample storage, sampling protocol, method of data analysis, type of disease, sample matrix, and potentially to clinical and disease factors. The results of this meta-analysis indicate that micro-chip FAIMS is a promising candidate for disease screening at the point-of-care, particularly for gastroenterology diseases. This review provides recommendations that should improve the techniques relevant to diagnostic accuracy of future VOC and point-of-care studies.

Relationship Between Serum Severe Acute Respiratory Syndrome Coronavirus 2 Nucleic Acid and Organ Damage in Coronavirus 2019 Patients: A Cohort Study

BACKGROUND

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) has spread worldwide and has the ability to damage multiple organs. However, information on serum SARS-CoV-2 nucleic acid (RNAemia) in patients affected by coronavirus disease 2019 (COVID-19) is limited.

METHODS

Patients who were admitted to Zhongnan Hospital of Wuhan University with laboratory-confirmed COVID-19 were tested for SARS-COV-2 RNA in serum from 28 January 2020 to 9 February 2020. Demographic data, laboratory and radiological findings, comorbidities, and outcomes data were collected and analyzed.

RESULTS

Eighty-five patients were included in the analysis. The viral load of throat swabs was significantly higher than of serum samples. The highest detection of SARS-CoV-2 RNA in serum samples was between 11 and 15 days after symptom onset. Analysis to compare patients with and without RNAemia provided evidence that computed tomography and some laboratory biomarkers (total protein, blood urea nitrogen, lactate dehydrogenase, hypersensitive troponin I, and D-dimer) were abnormal and that the extent of these abnormalities was generally higher in patients with RNAemia than in patients without RNAemia. Organ damage (respiratory failure, cardiac damage, renal damage, and coagulopathy) was more common in patients with RNAemia than in patients without RNAemia. Patients with vs without RNAemia had shorter durations from serum testing SARS-CoV-2 RNA. The mortality rate was higher among patients with vs without RNAemia.

CONCLUSIONS

In this study, we provide evidence to support that SARS-CoV-2 may have an important role in multiple organ damage. Our evidence suggests that RNAemia has a significant association with higher risk of in-hospital mortality.

Subchondral bone length in knee osteoarthritis: A deep learning derived imaging measure and its association with radiographic and clinical outcomes

OBJECTIVE

Develop a bone shape measure that reflects the extent of cartilage loss and bone flattening in knee osteoarthritis (OA) and test it against estimates of disease severity.

METHODS

A fast region-based convolutional neural network was trained to crop the knee joints in sagittal dual-echo steady state MRI sequences obtained from the Osteoarthritis Initiative (OAI). Publicly available annotations of the cartilage and menisci were used as references to annotate the tibia and the femur in 61 knees. Another deep neural network (U-Net) was developed to learn these annotations. Model predictions were compared with radiologist-driven annotations on an independent test set (27 knees). The U-Net was applied to automatically extract the knee joint structures on the larger OAI dataset (9,434 knees). We defined subchondral bone length (SBL), a novel shape measure characterizing the extent of overlying cartilage and bone flattening, and examined its relationship with radiographic joint space narrowing (JSN), concurrent WOMAC pain and disability as well as subsequent partial or total knee replacement (KR). Odds ratios for each outcome were estimated using relative changes in SBL on the OAI dataset into quartiles.

RESULT

Mean SBL values for knees with JSN were consistently different from knees without JSN. Greater changes of SBL from baseline were associated with greater pain and disability. For knees with medial or lateral JSN, the odds ratios between lowest and highest quartiles corresponding to SBL changes for future KR were 5.68 (95% CI:[3.90,8.27]) and 7.19 (95% CI:[3.71,13.95]), respectively.

CONCLUSION

SBL quantified OA status based on JSN severity. It has promise as an imaging marker in predicting clinical and structural OA outcomes.

Enhancing magnetic resonance imaging-driven Alzheimer's disease classification performance using generative adversarial learning

BACKGROUND

Generative adversarial networks (GAN) can produce images of improved quality but their ability to augment image-based classification is not fully explored. We evaluated if a modified GAN can learn from magnetic resonance imaging (MRI) scans of multiple magnetic field strengths to enhance Alzheimer's disease (AD) classification performance.

METHODS

T1-weighted brain MRI scans from 151 participants of the Alzheimer's Disease Neuroimaging Initiative (ADNI), who underwent both 1.5-Tesla (1.5-T) and 3-Tesla imaging at the same time were selected to construct a GAN model. This model was trained along with a three-dimensional fully convolutional network (FCN) using the generated images (3T*) as inputs to predict AD status. Quality of the generated images was evaluated using signal to noise ratio (SNR), Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and Natural Image Quality Evaluator (NIQE). Cases from the Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing (AIBL, n = 107) and the National Alzheimer's Coordinating Center (NACC, n = 565) were used for model validation.

RESULTS

The 3T*-based FCN classifier performed better than the FCN model trained using the 1.5-T scans. Specifically, the mean area under curve increased from 0.907 to 0.932, from 0.934 to 0.940, and from 0.870 to 0.907 on the ADNI test, AIBL, and NACC datasets, respectively. Additionally, we found that the mean quality of the generated (3T*) images was consistently higher than the 1.5-T images, as measured using SNR, BRISQUE, and NIQE on the validation datasets.

CONCLUSION

This study demonstrates a proof of principle that GAN frameworks can be constructed to augment AD classification performance and improve image quality.