Detection of Mild Cognitive Impairment Through Natural Language and Touchscreen Typing Processing

Abstract

Mild cognitive impairment (MCI), an identified prodromal stage of Alzheimer's Disease (AD), often evades detection in the early stages of the condition, when existing diagnostic methods are employed in the clinical setting. From an alternative perspective, smartphone interaction behavioral data, unobtrusively acquired in a non-clinical setting, can assist the screening and monitoring of MCI and its symptoms' progression. In this vein, the diagnostic ability of digital biomarkers, drawn from Fine Motor Impairment (FMI)-and Spontaneous Written Speech (SWS)-related data analysis, are examined here. In particular, keystroke dynamics derived from touchscreen typing activities, using Convolutional Neural Networks, along with linguistic features of SWS through Natural Language Processing (NLP), were used to distinguish amongst MCI patients and healthy controls (HC). Analytically, three indices of FMI (rigidity, bradykinesia and alternate finger tapping) and nine NLP features, related with lexical richness, grammatical, syntactical complexity, and word deficits, formed the feature space. The proposed approach was tested on two demographically matched groups of 11 MCI patients and 12 HC, having undergone the same neuropsychological tests, producing 4,930 typing sessions and 78 short texts, within 6 months, for analysis. A cascaded-classifier scheme was realized under three different feature combinations and validated via a Leave-One-Subject-Out cross-validation scheme. The acquired results have shown: (a) keystroke features with a k-NN classifier achieved an Area Under Curve (AUC) of 0.78 [95% confidence interval (CI):0.68-0.88; specificity/sensitivity (SP/SE): 0.64/0.92], (b) NLP features with a Logistic regression classifier achieved an AUC of 0.76 (95% CI: 0.65-0.85; SP/SE: 0.80/0.71), and (c) an ensemble model with the fusion of keystroke and NLP features resulted in AUC of 0.75 (95% CI:0.63-0.86; SP/SE 0.90/0.60). The current findings indicate the potentiality of new digital biomarkers to capture early stages of cognitive decline, providing a highly specific remote screening tool in-the-wild.

Diagnostic accuracy of keystroke dynamics as digital biomarkers for fine motor decline in neuropsychiatric disorders: a systematic review and meta-analysis

“…accelerometer displacement, IKD, Backspace ratio, avg. session length, number of sessions, circadian baseline similarity Mixed effects regression NA Mood Challenge for Research kit 1R01MH120168 Stange et al (2018) 57 Bipolar disorder In-the-wild through “BiAffect” smartphone application (10 weeks) 18 (NR) NA Hybrid (clinical assessment: HDRS, YMRS/ecological momentary assessment) Keyboard meta-data Subject level Root mean square successive difference (rMSSD) between keystrokes Multi-level and boot-strapped mediation analysis NA Mood Challenge for Research kit 1R01MH120168 Vizer et al (2015) 49 MCI In-the-clinic (4 typing sessions, 20–45 min each) 17 (81.12, 6, NR) 20 (79.24, 6, NR) Clinical evaluation: mini mental state examination (MMSE) Keystroke timing data and their linguistic content Subject level Paralinguistic: pause rate and duration, time per key and keystroke rate and linguistic features: sentence complexity, rate of nouns, verbs and adjectives NA Logistic regression US National Science Foundation graduate research fellowship, and the US National Library of Medicine Biomedical and Health Informatics Training Program at the University of Washington (grant number T15LM007442) Ntracha et al (2020) 50 MCI In-the-wild (6 months) 11 (67.2, 5.96, 81.8%) 12 (66.2, 4.72, 58.3%) Clinical Assessment (SCI, MMSE, FUCAS, FRSSD) KD and texts simulating Spontaneous Written Speech (SWS) Subject level NLP features and R/B/AFT indices from KD NA kNN (KD alone, logistic regression (NLP alone), ensemble model (fused features) Horizon 2020 research and innovation programme under grant agreement No 690494—i-PROGNOSIS Matarazzo et al (2019) 34 …”

“…To this end, Vizer and colleagues combined keystroke timing features including the HT and the pause rate with linguistic features collected in clinical settings to distinguish PreMCI subjects from age matched healthy controls 49 . Taking the analysis a step further, with the advancement in Natural Language Processing (NLP), and the capability of capturing objective linguistic features, usually not recognized by human raters, Ntracha et al employed NLP of Spontaneous Written Speech (SWS), fused with keystroke dynamics features captured in-the-wild, to reinforce the interplay of cognitive and fine motor functions 50 . Furthermore, the pronounced advancement in computational modeling now allows aligning multiple data lines, what facilitated the development of “behaviorgrams” that capture activity levels, physiological and behavioral signals on a longitudinal bases, yielding a more comprehensive overview of individuals’ health, yet without solid interpretability on longitudinal transient behavior 51 .…”

Perioperative Brain Health in the Older Adult: A Patient Safety Imperative

“…While current forms of cognitive screening before surgical care are useful for perioperative interventions, new technologies, such as natural language processing and automated speech analysis, [82][83][84] as well as gait evaluation technology, 85 offer low-cost and pragmatic approaches to risk stratification. One of the advantages of speech analysis is that, by using voice-recognition software, risk prediction can be done virtually.…”

Secondary Prevention of Dementia: Combining Risk Factors and Scalable Screening Technology

“…These data on fine motor control, language abilities and processing speed, can be used to build predictive models for early disease detection. For example, Ntracha et al ( 139 ) used touchscreen typing characteristics (participants were asked to type stories on the phone) to build a model with diagnostic ability that distinguishes MCI patients from controls. The best performing model had accuracy of 80% (AUC = 0.75), which is in a similar range to many other dementia prediction models that use large cohort data ( 140 , 141 ).…”