MicroRNA classification and discovery for major depressive disorder diagnosis: Towards a robust and interpretable machine learning approach.

BACKGROUND: Major depressive disorder (MDD) is notably underdiagnosed and undertreated due to its complex nature and subjective diagnostic methods. Biomarker identification would help provide a clearer understanding of MDD aetiology. Although machine learning (ML) has been implemented in previous studies to study the alteration of microRNA (miRNA) levels in MDD cases, clinical translation has not been feasible due to the lack of interpretability (i.e. too many miRNAs for consideration) and stability. METHODS: This study applied logistic regression (LR) model to the blood miRNA expression profile to differentiate patients with MDD (n = 60) from healthy controls (HCs, n = 60). Embedded (L1-regularised logistic regression) feature selector was utilised to extract clinically relevant miRNAs, and optimized for clinical application. RESULTS: Patients with MDD could be differentiated from HCs with the area under the receiver operating characteristic curve (AUC) of 0.81 on testing data when all available miRNAs were considered (which served as a benchmark). Our LR model selected miRNAs up to 5 (known as LR-5 model) emerged as the best model because it achieved a moderate classification ability (AUC = 0.75), relatively high interpretability (feature number = 5) and stability (ϕ̂Z=0.55) compared to the benchmark. The top-ranking miRNAs identified by our model have demonstrated associations with MDD pathways involving cytokine signalling in the immune system, the reelin signalling pathway, programmed cell death and cellular responses to stress. CONCLUSION: The LR-5 model, which is optimised based on ML design factors, may lead to a robust and clinically usable MDD diagnostic tool.

Using an Interpretable Amino Acid-Based Machine Learning Method to Enhance the Diagnosis of Major Depressive Disorder.

Background: Major depressive disorder (MDD) is a leading cause of disability worldwide. At present, however, there are no established biomarkers that have been validated for diagnosing and treating MDD. This study sought to assess the diagnostic and predictive potential of the differences in serum amino acid concentration levels between MDD patients and healthy controls (HCs), integrating them into interpretable machine learning models. Methods: In total, 70 MDD patients and 70 HCs matched in age, gender, and ethnicity were recruited for the study. Serum amino acid profiling was conducted by means of chromatography-mass spectrometry. A total of 21 metabolites were analysed, with 17 from a preset amino acid panel and the remaining 4 from a preset kynurenine panel. Logistic regression was applied to differentiate MDD patients from HCs. Results: The best-performing model utilised both feature selection and hyperparameter optimisation and yielded a moderate area under the receiver operating curve (AUC) classification value of 0.76 on the testing data. The top five metabolites identified as potential biomarkers for MDD were 3-hydroxy-kynurenine, valine, kynurenine, glutamic acid, and xanthurenic acid. Conclusions: Our study highlights the potential of using an interpretable machine learning analysis model based on amino acids to aid and increase the diagnostic accuracy of MDD in clinical practice.

Optimizing Mental Workload Estimation by Detecting Baseline State Using Vector Phase Analysis Approach.

Improper baseline return from the previous task-evoked hemodynamic response (HR) can contribute to a large variation in the subsequent HR, affecting the estimation of mental workload in brain-computer interface systems. In this study, we proposed a method using vector phase analysis to detect the baseline state as being optimal or suboptimal. We hypothesize that selecting neuronal-related HR as observed in the optimal-baseline blocks can lead to an improvement in estimating mental workload. Oxygenated and deoxygenated hemoglobin concentration changes were integrated as parts of the vector phase. The proposed method was applied to a block-design functional near-infrared spectroscopy dataset (total blocks = 1384), measured on 24 subjects performing multiple difficulty levels of mental arithmetic task. Significant differences in hemodynamic signal change were observed between the optimal- and suboptimal-baseline blocks detected using the proposed method. This supports the effectiveness of the proposed method in detecting baseline state for better estimation of mental workload. The results further highlight the need of customized recovery duration. In short, the proposed method offers a practical approach to detect task-evoked signals, without the need of extra probes.

fNIRS-based functional connectivity estimation using semi-metric analysis to study decision making by nursing students and registered nurses.

This study aims to investigate the generalizability of the semi-metric analysis of the functional connectivity (FC) for functional near-infrared spectroscopy (fNIRS) by applying it to detect the dichotomy in differential FC under affective and neutral emotional states in nursing students and registered nurses during decision making. The proposed method employs wavelet transform coherence to construct FC networks and explores semi-metric analysis to extract network redundancy features, which has not been considered in conventional fNIRS-based FC analyses. The trials of the proposed method were performed on 19 nursing students and 19 registered nurses via a decision-making task under different emotional states induced by affective and neutral emotional stimuli. The cognitive activities were recorded using fNIRS, and the emotional stimuli were adopted from the International Affective Digitized Sound System (IADS). The induction of emotional effects was validated by heart rate variability (HRV) analysis. The experimental results by the proposed method showed significant difference (FDR-adjusted p = 0.004) in the nursing students' cognitive FC network under the two different emotional conditions, and the semi-metric percentage (SMP) of the right prefrontal cortex (PFC) was found to be significantly higher than the left PFC (FDR-adjusted p = 0.036). The benchmark method (a typical weighted graph theory analysis) gave no significant results. In essence, the results support that the semi-metric analysis can be generalized and extended to fNIRS-based functional connectivity estimation.

Characterizing Functional Near Infrared Spectroscopy (fNIRS)-based Connectivity as Cost-effective Small World Network using Orthogonal Minimal Spanning Trees (OMSTs).

This paper reported data-driven functional connectivity (FC) analytical method to investigate functional near infrared spectroscopy (fNIRS)-based connectivity. We evaluated the synchronization of oxygenated hemoglobin using Pearson's correlation and employed orthogonal minimal spanning trees (OMSTs) in characterizing brain connectivity. Then we compared the resultant global cost efficiency and robustness with those generated by non-human i.e. lattice and random networks. We also further benchmarked our method using proportional threshold. Results from 59 healthy subjects demonstrated global cost efficiency and assortativity varied in lattice and random network significantly (p < 0.05), highlighting the potential of OMSTs in extracting true neuronal network. Moreover, the inadequate of proportional threshold in extracting small world network from the same dataset supported that the OMSTs might be the better alternative in FC analysis especially in evaluating cost-efficiency and robustness of network.

A Unified Analytical Framework With Multiple fNIRS Features for Mental Workload Assessment in the Prefrontal Cortex.

Knowing the actual level of mental workload is important to ensure the efficacy of brain-computer interface (BCI) based cognitive training. Extracting signals from limited area of a brain region might not reveal the actual information. In this study, a functional near-infrared spectroscopy (fNIRS) device equipped with multi-channel and multi-distance measurement capability was employed for the development of an analytical framework to assess mental workload in the prefrontal cortex (PFC). In addition to the conventional features, e.g. hemodynamic slope, we introduced a new feature - deep contribution ratio which is the proportion of cerebral hemodynamics to the fNIRS signals. Multiple sets of features were examined by a simple logical operator to suppress the false detection rate in identifying the activated channels. Using the number of activated channels as input to a linear support vector machine (SVM), the performance of the proposed analytical framework was assessed in classifying three levels of mental workload. The best set of features involves the combination of hemodynamic slope and deep contribution ratio, where the identified number of activated channels returned an average accuracy of 80.6% in predicting mental workload, compared to a single conventional feature (accuracy: 59.8%). This suggests the feasibility of the proposed analytical framework with multiple features as a means towards a more accurate assessment of mental workload in fNIRS-based BCI applications.

Automated Thresholding Method for fNIRS-Based Functional Connectivity Analysis: Validation With a Case Study on Alzheimer's Disease.

While functional integration has been suggested to reflect brain health, non-standardized network thresholding methods complicate network interpretation. We propose a new method to analyze functional near-infrared spectroscopy-based functional connectivity (fNIRS-FC). In this study, we employed wavelet analysis for motion correction and orthogonal minimal spanning trees (OMSTs) to derive the brain connectivity. The proposed method was applied to an Alzheimer's disease (AD) dataset and was compared with a number of well-known thresholding techniques. The results demonstrated that the proposed method outperformed the benchmarks in filtering cost-effective networks and in differentiation between patients with mild AD and healthy controls. The results also supported the proposed method as a feasible technique to analyze fNIRS-FC, especially with cost-efficiency, assortativity and laterality as a set of effective features for the diagnosis of AD.

Denoising of neuronal signal from mixed systemic low-frequency oscillation using peripheral measurement as noise regressor in near-infrared imaging.

Functional near-infrared spectroscopy (fNIRS) is a noninvasive functional imaging technique measuring hemodynamic changes including oxygenated ( O 2 Hb ) and deoxygenated (HHb) hemoglobin. Low frequency (LF; 0.01 to 0.15 Hz) band is commonly analyzed in fNIRS to represent neuronal activation. However, systemic physiological artifacts (i.e., nonneuronal) likely occur also in overlapping frequency bands. We measured peripheral photoplethysmogram (PPG) signal concurrently with fNIRS (at prefrontal region) to extract the low-frequency oscillations (LFOs) as systemic noise regressors. We investigated three main points in this study: (1) the relationship between prefrontal fNIRS and peripheral PPG signals; (2) the denoising potential using these peripheral LFOs, and (3) the innovative ways to avoid the false-positive result in fNIRS studies. We employed spatial working memory (WM) and control tasks (e.g., resting state) to illustrate these points. Our results showed: (1) correlation between signals from prefrontal fNIRS and peripheral PPG is region-dependent. The high correlation with peripheral ear signal (i.e., O 2 Hb ) occurred mainly in frontopolar regions in both spatial WM and control tasks. This may indicate the finding of task-dependent effect even in peripheral signals. We also found that the PPG recording at the ear has a high correlation with prefrontal fNIRS signal than the finger signals. (2) The systemic noise was reduced by 25% to 34% on average across regions, with a maximum of 39% to 58% in the highly correlated frontopolar region, by using these peripheral LFOs as noise regressors. (3) By performing the control tasks, we confirmed that the statistically significant activation was observed in the spatial WM task, not in the controls. This suggested that systemic (and any other) noises unlikely violated the major statistical inference. (4) Lastly, by denoising using the task-related signals, the significant activation of region-of-interest was still observed suggesting the manifest task-evoked response in the spatial WM task.

Functional Connectivity Analysis on Mild Alzheimer's Disease, Mild Cognitive Impairment and Normal Aging using fNIRS.

This paper reports a functional connectivity analysis at prefrontal cortex (PFC) during semantic verbal fluency task (SVFT) for three groups of elderly people, i.e., normal aging (NA), mild cognitive impairment (MCI) and mild Alzheimer's disease (AD). Functional Near Infrared Spectroscopy (fNIRS) was used to measure neuronal activities. A new software algorithm was developed to process fNIRS signals and to derive the parameters of functional connectivity. The synchronization of oxygenated hemoglobin signals from paired channels was evaluated using their temporal correlation. Results from 61 subjects of experiment show that a general decline in functional connectivity from NA (edge count $=$ 307) to AD (edge count $=$170), and the laterality between left and right PFC became insignificant $( \mathrm {p}>0.01)$ at AD stage. Moreover, the NA group demonstrated a significantly higher clustering coefficient than the AD group $( \mathrm {p}< 0.01)$, indicating the NA has higher regularity in brain network. Using semantic verbal fluency task, this work demonstrated fNIRS as a feasible measuring instrument to differentiate AD from NA based on functional connectivity, with clustering coefficient and laterality as suitable biomarkers.