In JMIR mHealth and uHealth

BACKGROUND : Stress is an important predictor of mental health problems such as burnout and depression. Acute stress is considered adaptive, whereas chronic stress is viewed as detrimental to well-being. To aid in the early detection of chronic stress, machine learning models are increasingly trained to learn the quantitative relation from digital footprints to self-reported stress. Prior studies have investigated general principles in population-wide studies, but the extent to which the findings apply to individuals is understudied.

OBJECTIVE : We aimed to explore to what extent machine learning models can leverage features of smartphone app use log data to recognize momentary subjective stress in individuals, which of these features are most important for predicting stress and represent potential digital markers of stress, the nature of the relations between these digital markers and stress, and the degree to which these relations differ across people.

METHODS : Student participants (N=224) self-reported momentary subjective stress 5 times per day up to 60 days in total (44,381 observations); in parallel, dedicated smartphone software continuously logged their smartphone app use. We extracted features from the log data (eg, time spent on app categories such as messenger apps and proxies for sleep duration and onset) and trained machine learning models to predict momentary subjective stress from these features using 2 approaches: modeling general relations at the group level (nomothetic approach) and modeling relations for each person separately (idiographic approach). To identify potential digital markers of momentary subjective stress, we applied explainable artificial intelligence methodology (ie, Shapley additive explanations). We evaluated model accuracy on a person-to-person basis in out-of-sample observations.

RESULTS : We identified prolonged use of messenger and social network site apps and proxies for sleep duration and onset as the most important features across modeling approaches (nomothetic vs idiographic). The relations of these digital markers with momentary subjective stress differed from person to person, as did model accuracy. Sleep proxies, messenger, and social network use were heterogeneously related to stress (ie, negative in some and positive or zero in others). Model predictions correlated positively and statistically significantly with self-reported stress in most individuals (median person-specific correlation=0.15-0.19 for nomothetic models and median person-specific correlation=0.00-0.09 for idiographic models).

CONCLUSIONS : Our findings indicate that smartphone log data can be used for identifying digital markers of stress and also show that the relation between specific digital markers and stress differs from person to person. These findings warrant follow-up studies in other populations (eg, professionals and clinical populations) and pave the way for similar research using physiological measures of stress.

Aalbers George, Hendrickson Andrew T, Vanden Abeele Mariek Mp, Keijsers Loes

2023-Mar-23

digital biomarker, digital phenotype, machine learning, mobile health, mobile phone, personalized models

In Journal of medical Internet research ; h5-index 88.0

BACKGROUND : Assessing a patient's suicide risk is challenging for health professionals because it depends on voluntary disclosure by the patient and often has limited resources. The application of novel machine learning approaches to determine suicide risk has clinical utility.

OBJECTIVE : This study aimed to investigate cross-sectional and longitudinal approaches to assess suicidality based on acoustic voice features of psychiatric patients using artificial intelligence.

METHODS : We collected 348 voice recordings during clinical interviews of 104 patients diagnosed with mood disorders at baseline and 2, 4, 8, and 12 months after recruitment. Suicidality was assessed using the Beck Scale for Suicidal Ideation and suicidal behavior using the Columbia Suicide Severity Rating Scale. The acoustic features of the voice, including temporal, formal, and spectral features, were extracted from the recordings. A between-person classification model that examines the vocal characteristics of individuals cross sectionally to detect individuals at high risk for suicide and a within-person classification model that detects considerable worsening of suicidality based on changes in acoustic features within an individual were developed and compared. Internal validation was performed using 10-fold cross validation of audio data from baseline to 2-month and external validation was performed using data from 2 to 4 months.

RESULTS : A combined set of 12 acoustic features and 3 demographic variables (age, sex, and past suicide attempts) were included in the single-layer artificial neural network for the between-person classification model. Furthermore, 13 acoustic features were included in the extreme gradient boosting machine learning algorithm for the within-person model. The between-person classifier was able to detect high suicidality with 69% accuracy (sensitivity 74%, specificity 62%, area under the receiver operating characteristic curve 0.62), whereas the within-person model was able to predict worsening suicidality over 2 months with 79% accuracy (sensitivity 68%, specificity 84%, area under receiver operating characteristic curve 0.67). The second model showed 62% accuracy in predicting increased suicidality in external sets.

CONCLUSIONS : Within-person analysis using changes in acoustic features within an individual is a promising approach to detect increased suicidality. Automated analysis of voice can be used to support the real-time assessment of suicide risk in primary care or telemedicine.

Min Sooyeon, Shin Daun, Rhee Sang Jin, Park C Hyung Keun, Yang Jeong Hun, Song Yoojin, Kim Min Ji, Kim Kyungdo, Cho Won Ik, Kwon Oh Chul, Ahn Yong Min, Lee Hyunju

2023-Mar-23

artificial intelligence, digital health tool, mood disorder, prediction, screening, suicide, voice analysis

In Soft robotics

Jin Tao, Wang Tianhong, Xiong Quan, Tian Yingzhong, Li Long, Zhang Quan, Yeow Chen-Hua

2023-Mar-23

modular actuation, modular soft robot, origami, soft pneumatic actuators

In Journal of magnetic resonance imaging : JMRI

BACKGROUND : Multidelay arterial spin labeling (ASL) generates time-resolved perfusion maps, which may provide sufficient and accurate hemodynamic information in carotid stenosis.

PURPOSE : To use imaging markers derived from multidelay ASL magnetic resonance imaging (MRI) and to determine the optimal strategy for predicting cerebral hyperperfusion after carotid endarterectomy (CEA).

STUDY TYPE : Prospective observational cohort.

SUBJECTS : A total of 79 patients who underwent CEA for carotid stenosis.

FIELD STRENGTH/SEQUENCE : A 3.0 T/pseudo-continuous ASL with three postlabeling delays of 1.0, 1.57, and 2.46 seconds using fast-spin echo readout.

ASSESSMENT : Cerebral perfusion pressure, antegrade, and collateral flow were scored on a four-grade ordinal scale based on preoperative multidelay ASL perfusion maps. Simultaneously, quantitative hemodynamic parameters including cerebral blood flow (CBF), arterial transit time (ATT), relative CBF (rCBF) and relative ATT (rATT; ipsilateral/contralateral values) were calculated. On the CBF ratio map obtained through dividing postoperative by preoperative CBF map, regions of interest were placed covering ipsilateral middle cerebral artery territory. Three neuroradiologists conducted this procedure. Cerebral hyperperfusion was defined as a CBF ratio >2.

STATISTICAL TESTS : Weighted κ values, independent sample t test, chi-square test, Mann-Whitney U-test, multivariable logistic regression analysis, receiver-operating characteristic curve analysis, and Delong test. Significance level was P < 0.05.

RESULTS : Cerebral hyperperfusion was observed in 15 (19%) patients. Higher blood pressure (odd ratio [OR] = 1.08) and carotid near-occlusion (NO; OR = 7.31) were clinical risk factors for postoperative hyperperfusion. Poor ASL perfusion score (OR = 37.33), decreased CBF (OR = 0.74), prolonged ATT (OR = 1.02), lower rCBF (OR = 0.91), and higher rATT (OR = 1.12) were independent imaging predictors of hyperperfusion. ASL perfusion score exhibited the highest specificity (95.3%), while CBF exhibited the highest sensitivity (93.3%) for the prediction of hyperperfusion. When combined with ASL perfusion score, CBF and ATT, the predictive ability was significantly higher than using blood pressure and NO alone (AUC: 0.98 vs. 0.78).

DATA CONCLUSIONS : Multidelay ASL can accurately predict cerebral hyperperfusion after CEA with high sensitivity and specificity.

EVIDENCE LEVEL : 2 TECHNICAL EFFICACY: Stage 5.

Fan Xiaoyuan, Lai Zhichao, Lin Tianye, Li Kang, Hou Bo, You Hui, Wei Juan, Qu Jianxun, Liu Bao, Zuo Zhentao, Feng Feng

2023-Mar-23

arterial spin labeling, carotid endarterectomy, carotid stenosis, cerebral hyperperfusion syndrome

In Computer methods in biomechanics and biomedical engineering

Saraswat Mala, Dubey Anil Kumar

2023-Mar-23

Time-series data classification, brain–computer interface, electroencephalography signal, enhanced bi-directional long short-term memory, ensemble feature extraction, probability ratio-based reptile search algorithm

In Turk patoloji dergisi

Basak Kayhan, Ozyoruk Kutsev Bengisu, Demir Derya

2023-Mar-23