Wearable data from subjects playing Super Mario, sitting university exams, or performing physical exercise help detect acute mood episodes via self-supervised learning DOI Creative Commons
Filippo Corponi, Bryan M. Li, Gerard Anmella

et al.

arXiv (Cornell University), Journal Year: 2023, Volume and Issue: unknown

Published: Jan. 1, 2023

Personal sensing, leveraging data passively and near-continuously collected with wearables from patients in their ecological environment, is a promising paradigm to monitor mood disorders (MDs), major determinant of worldwide disease burden. However, collecting annotating wearable very resource-intensive. Studies this kind can thus typically afford recruit only couple dozens patients. This constitutes one the obstacles applying modern supervised machine learning techniques MDs detection. In paper, we overcome bottleneck advance detection acute episode vs stable state on back recent advances self-supervised (SSL). leverages unlabelled learn representations during pre-training, subsequently exploited for task. First, open-access datasets recording an Empatica E4 spanning different, unrelated MD monitoring, personal sensing tasks -- emotion recognition Super Mario players stress undergraduates devised pre-processing pipeline performing on-/off-body detection, sleep-wake segmentation, (optionally) feature extraction. With 161 E4-recorded subjects, introduce E4SelfLearning, largest date open access collection, its pipeline. Second, show that SSL confidently outperforms fully-supervised pipelines using either our novel E4-tailored Transformer architecture (E4mer) or classical baseline XGBoost: 81.23% against 75.35% 72.02% (XGBoost) correctly classified segments 64 (half acute, half stable) Lastly, illustrate performance strongly associated specific surrogate task employed pre-training as well availability.

Language: Английский

Exploring Digital Biomarkers of Illness Activity in Mood Episodes: Hypotheses Generating and Model Development Study DOI Creative Commons
Gerard Anmella, Filippo Corponi, Bryan M. Li

et al.

JMIR mhealth and uhealth, Journal Year: 2023, Volume and Issue: 11, P. e45405 - e45405

Published: March 20, 2023

Depressive and manic episodes within bipolar disorder (BD) major depressive (MDD) involve altered mood, sleep, activity, alongside physiological alterations wearables can capture. Firstly, we explored whether wearable data could predict (aim 1) the severity of an acute affective episode at intra-individual level 2) polarity euthymia among different individuals. Secondarily, which were related to prior predictions, generalization across patients, associations between symptoms data. We conducted a prospective exploratory observational study including patients with BD MDD on (manic, depressed, mixed) whose recorded using research-grade (Empatica E4) 3 consecutive time points (acute, response, remission episode). Euthymic healthy controls during single session (approximately 48 h). Manic assessed standardized psychometric scales. Physiological included following channels: acceleration (ACC), skin temperature, blood volume pulse, heart rate (HR), electrodermal activity (EDA). Invalid removed rule-based filter, channels aligned 1-second units segmented window lengths 32 seconds, as best-performing parameters. developed deep learning predictive models, channels' individual contribution permutation feature importance analysis, computed scales' items normalized mutual information (NMI). present novel, fully automated method for preprocessing analysis from device, viable supervised pipeline time-series analyses. Overall, 35 sessions (1512 hours) 12 mixed, euthymic) 7 (mean age 39.7, SD 12.6 years; 6/19, 32% female) analyzed. The mood was predicted moderate (62%-85%) accuracies 1), their (70%) accuracy 2). most relevant features former tasks ACC, EDA, HR. There fair agreement in classification (Kendall W=0.383). Generalization models unseen overall low accuracy, except models. ACC associated "increased motor activity" (NMI>0.55), "insomnia" (NMI=0.6), "motor inhibition" (NMI=0.75). EDA "aggressive behavior" (NMI=1.0) "psychic anxiety" (NMI=0.52). show potential identify specific mania depression quantitatively, both MDD. Motor stress-related (EDA HR) stand out digital biomarkers predicting depression, respectively. These findings represent promising pathway toward personalized psychiatry, allow early identification intervention episodes.

Language: Английский

Citations

19
Automated mood disorder symptoms monitoring from multivariate time-series sensory data: getting the full picture beyond a single number DOI Creative Commons
Filippo Corponi, Bryan M. Li, Gerard Anmella

et al.

Translational Psychiatry, Journal Year: 2024, Volume and Issue: 14(1)

Published: March 26, 2024

Mood disorders (MDs) are among the leading causes of disease burden worldwide. Limited specialized care availability remains a major bottleneck thus hindering pre-emptive interventions. MDs manifest with changes in mood, sleep, and motor activity, observable ecological physiological recordings thanks to recent advances wearable technology. Therefore, near-continuous passive collection data from wearables daily life, analyzable machine learning (ML), could mitigate this problem, bringing monitoring outside clinician's office. Previous works predict single label, either state or psychometric scale total score. However, clinical practice suggests that same label may underlie different symptom profiles, requiring specific treatments. Here we bridge gap by proposing new task: inferring all items HDRS YMRS, two most widely used standardized scales for assessing symptoms, using wearables. To end, develop deep pipeline score symptoms large cohort MD patients show agreement between predictions assessments an expert clinician is clinically significant (quadratic Cohen's κ macro-average F1 both 0.609). While doing so, investigate several solutions ML challenges associated task, including multi-task learning, class imbalance, ordinal target variables, subject-invariant representations. Lastly, illustrate importance testing on out-of-distribution samples.

Language: Английский

Citations

6
Omics approaches open new horizons in major depressive disorder: from biomarkers to precision medicine DOI Creative Commons

Fabiola Stolfi,

Hugo Abreu,

Riccardo Sinella

et al.

Frontiers in Psychiatry, Journal Year: 2024, Volume and Issue: 15

Published: June 13, 2024

Major depressive disorder (MDD) is a recurrent episodic mood that represents the third leading cause of disability worldwide. In MDD, several factors can simultaneously contribute to its development, which complicates diagnosis. According practical guidelines, antidepressants are first-line treatment for moderate severe major episodes. Traditional strategies often follow one-size-fits-all approach, resulting in suboptimal outcomes many patients who fail experience response or recovery and develop so-called “therapy-resistant depression”. The high biological clinical inter-variability within lack robust biomarkers hinder finding specific therapeutic targets, contributing failure rates. this frame, precision medicine, paradigm tailors medical interventions individual characteristics, would help allocate most adequate effective each patient while minimizing side effects. particular, multi-omic studies may unveil intricate interplays between genetic predispositions exposure environmental through study epigenomics, transcriptomics, proteomics, metabolomics, gut microbiomics, immunomics. integration flow information into molecular pathways produce better than current psychopharmacological targets singular mainly related monoamine systems, disregarding complex network our organism. concept system biomedicine involves analysis enormous datasets generated with different technologies, creating “patient fingerprint”, defines underlying mechanisms every patient. This review, centered on explores approaches as tools prediction MDD at single-patient level. It investigates how combining existing technologies used diagnostic, stratification, prognostic, treatment-response discovery artificial intelligence improve assessment MDD.

Language: Английский

Citations

6
Predicting the severity of mood and neuropsychiatric symptoms from digital biomarkers using wearable physiological data and deep learning DOI Creative Commons
Yuri Rykov, Kok Pin Ng, Michael D. Patterson

et al.

Computers in Biology and Medicine, Journal Year: 2024, Volume and Issue: 180, P. 108959 - 108959

Published: July 31, 2024

Neuropsychiatric symptoms (NPS) and mood disorders are common in individuals with mild cognitive impairment (MCI) increase the risk of progression to dementia. Wearable devices collecting physiological behavioral data can help remote, passive, continuous monitoring moods NPS, overcoming limitations inconveniences current assessment methods. In this longitudinal study, we examined predictive ability digital biomarkers based on sensor from a wrist-worn wearable determine severity NPS daily basis older adults predominant MCI. addition conventional biomarkers, such as heart rate variability skin conductance levels, leveraged deep-learning features derived using self-supervised convolutional autoencoder. Models combining deep predicted depression scores correlation r = 0.73 average, total disorder 0.67, 0.69 study population. Our findings demonstrated potential collected wearables learning methods be used for unobtrusive assessments mental health adults, including those TRIAL REGISTRATION: This trial was registered ClinicalTrials.gov (NCT05059353) September 28, 2021, titled "Effectiveness Safety Digitally Based Multidomain Intervention Mild Cognitive Impairment".

Language: Английский

Citations

4
Machine learning applied to digital phenotyping: A systematic literature review and taxonomy DOI
Marília Pit dos Santos, Wesllei Felipe Heckler, Rodrigo Simon Bavaresco

et al.

Computers in Human Behavior, Journal Year: 2024, Volume and Issue: 161, P. 108422 - 108422

Published: Aug. 24, 2024

Language: Английский

Citations

4
Utilizing Digital Phenotyping to Discriminate Unipolar Depression and Bipolar Disorder: A Systematic Review (Preprint) DOI Creative Commons

Rongrong Zhong,

Xiaohui Wu, Jun Chen

et al.

Journal of Medical Internet Research, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 18, 2025

Language: Английский

Citations

0
Bipolar disorders: an update on critical aspects DOI Creative Commons
Vincenzo Oliva, Giovanna Fico, Michele De Prisco

et al.

The Lancet Regional Health - Europe, Journal Year: 2024, Volume and Issue: 48, P. 101135 - 101135

Published: Nov. 29, 2024

Language: Английский

Citations

3
Machine Learning as a Tool to Find New Pharmacological Targets in Mood Disorders: A Systematic Review DOI Creative Commons

Joana Romão,

António Melo,

Rita André

et al.

Current Treatment Options in Psychiatry, Journal Year: 2024, Volume and Issue: 11(3), P. 241 - 264

Published: Aug. 2, 2024

Abstract Purpose of Review Mood disorders (MD) are mental that need accurate diagnosis and proper treatment. Growing volume data from neurobehavioral sciences is becoming complex for traditional research to analyze. New drugs’ slow development fails meet the needs disorders. Machine Learning (ML) techniques support by refining detection, diagnosis, treatment, research, being employed expedite discovery pharmacological targets. This review aims assess evidence regarding contribution ML in finding new targets adults with MD. Recent findings The most significant area amongst MD major depressive disorder. identified target gene candidates, pathways biomarkers related MD, which can pave way promising therapeutic strategies. was also found enhance diagnostic accuracy. Summary have potential bridge gap between biological chemical drug information, providing discoveries agents.

Language: Английский

Citations

2
Automated mood disorder symptoms monitoring from multivariate time-series sensory data: Getting the full picture beyond a single number DOI Creative Commons
Filippo Corponi, Bryan M. Li, Gerard Anmella

et al.

medRxiv (Cold Spring Harbor Laboratory), Journal Year: 2023, Volume and Issue: unknown

Published: March 29, 2023

Abstract Mood disorders are among the leading causes of disease burden worldwide. They manifest with changes in mood, sleep, and motor-activity, observable physiological data. Despite effective treatments being available, limited specialized care availability is a major bottleneck, hindering preemptive interventions. Nearcontinuous passive collection data from wearables daily life, analyzable machine learning, could mitigate this problem, bringing mood monitoring outside doctor’s office. Previous works attempted predicting single label, e.g. state or psychometric scale total score. However, clinical practice suggests that same label can underlie different symptom profiles, requiring personalized treatment. In work we address limitation by proposing new task: inferring all items Hamilton Depression Rating Scale (HDRS) Young Mania (YMRS), most-widely used standardized questionnaires for assessing depression mania symptoms respectively, two polarities disorders. Using naturalistic, single-center cohort patients disorder (N=75), develop an artificial neural network (ANN) inputs wearable device scores on HDRS YMRS moderate agreement (quadratic Cohen’s κ = 0.609) assessments clinician. We also show that, when using as input recorded further away were collected clinician, ANN performance deteriorates, pointing to distribution shift, likely across both scales This task challenging research into domain-adaptation should be prioritized towards real-world implementations.

Language: Английский

Citations

5
Electrodermal activity in bipolar disorder: Differences between mood episodes and clinical remission using a wearable device in a real-world clinical setting DOI
Gerard Anmella, Ariadna Mas, Miriam Sanabra

et al.

Journal of Affective Disorders, Journal Year: 2023, Volume and Issue: 345, P. 43 - 50

Published: Oct. 21, 2023

Language: Английский

Citations

5