Machine learning approaches for integrating clinical and imaging features in late-life depression classification and response prediction

Challenges and prospects in bridging precision medicine and artificial intelligence in genomic psychiatric treatment

Precision medicine is transforming psychiatric treatment by tailoring personalized healthcare interventions based on clinical, genetic, environmental, and lifestyle factors to optimize medication management. This study investigates how artificial intelligence (AI) and machine learning (ML) can address key challenges in integrating pharmacogenomics (PGx) into psychiatric care. In this integration, AI analyzes vast genomic datasets to identify genetic markers linked to psychiatric conditions. AI-driven models integrating genomic, clinical, and demographic data demonstrated high accuracy in predicting treatment outcomes for major depressive disorder and bipolar disorder. This study also examines the pressing challenges and provides strategic directions for integrating AI and ML in genomic psychiatry, highlighting the importance of ethical considerations and the need for personalized treatment. Effective implementation of AI-driven clinical decision support systems within electronic health records is crucial for translating PGx into routine psychiatric care. Future research should focus on developing enhanced AI-driven predictive models, privacy-preserving data exchange, and robust informatics systems to optimize patient outcomes and advance precision medicine in psychiatry.

Models for depression recognition and efficacy assessment based on clinical and sequencing data

Major depression is a complex psychiatric disorder that includes genetic, neurological, and cognitive factors. Early detection and intervention can prevent progression, and help select the best treatment. Traditional clinical diagnosis tends to be subjective and misdiagnosed. Based on this, this study leverages clinical scale assessments and sequencing data to construct disease prediction models. Firstly, data undergoes preprocessing involving normalization and other requisite procedures. Feature engineering is then applied to curate subsets of features, culminating in the construction of a model through the implementation of machine learning and deep learning algorithms. In this study, 18 features with significant differences between patients and healthy controls were selected. The depression recognition model was constructed by deep learning with an accuracy of 87.26 % and an AUC of 91.56 %, which can effectively distinguish patients with depression from healthy controls. In addition, 33 features selected by recursive feature elimination method were used to construct a prognostic effect model of patients after 2 weeks of treatment, with an accuracy of 75.94 % and an AUC of 83.33 %. The results show that the deep learning algorithm based on clinical and sequencing data has good accuracy and provides an objective and accurate method for the diagnosis and pharmacodynamic prediction of depression. Furthermore, the selected differential features can serve as candidate biomarkers to provide valuable clues for diagnosis and efficacy prediction.

Neural activity during inhibitory control predicts suicidal ideation with machine learning

Suicide is a leading cause of death in the US and worldwide. Current strategies for preventing suicide are often focused on the identification and treatment of risk factors, especially suicidal ideation (SI). Hence, developing data-driven biomarkers of SI may be key for suicide prevention and intervention. Prior attempts at biomarker-based prediction models for SI have primarily used expensive neuroimaging technologies, yet clinically scalable and affordable biomarkers remain elusive. Here, we investigated the classification of SI using machine learning (ML) on a dataset of 76 subjects with and without SI(+/-) (n = 38 each), who completed a neuro-cognitive assessment session synchronized with electroencephalography (EEG). SI+/- groups were matched for age, sex, and mental health symptoms of depression and anxiety. EEG was recorded at rest and while subjects engaged in four cognitive tasks of inhibitory control, interference processing, working memory, and emotion bias. We parsed EEG signals in physiologically relevant theta (4-8 Hz), alpha (8-13 Hz), and beta (13-30 Hz) frequencies and performed cortical source imaging on the neural signals. These data served as SI predictors in ML models. The best ML model was obtained for beta band power during the inhibitory control (IC) task, demonstrating high sensitivity (89%), specificity (98%). Shapley explainer plots further showed top neural predictors as feedback-related power in the visual and posterior default mode networks and response-related power in the ventral attention, fronto-parietal, and sensory-motor networks. We further tested the external validity of the model in an independent clinically depressed sample (n = 35, 12 SI+) that engaged in an adaptive test version of the IC task, demonstrating 50% sensitivity and 61% specificity in this sample. Overall, the study suggests a promising, scalable EEG-based biomarker approach to predict SI that may serve as a target for risk identification and intervention.

NEFA can serve as good biological markers for the diagnosis of depression in adolescents

BACKGROUND

The incidence of adolescent depression has markedly risen in recent years, with a high recurrence rate into adulthood. Diagnosis in adolescents is challenging due to subjective factors, highlighting the crucial need for objective diagnostic markers.

METHODS

Our study enrolled 204 participants, including healthy controls (n = 88) and first-episode adolescent depression patients (n = 116). Serum samples underwent gas chromatography-mass spectrometry (GC-MS) analysis to assess non-esterified fatty acids (NEFA) expression. Machine learning and ROC analysis were employed to identify potential biomarkers, followed by bioinformatics analysis to explore underlying mechanisms.

RESULTS

Nearly all differentially expressed NEFA exhibited significant downregulation. Notably, nonanoic acid, cis-10-pentadecenoic acid, cis-10-carboenoic acid, and cis-11-eicosenoic acid demonstrated excellent performance in distinguishing adolescent depression patients. Metabolite-gene interaction analysis revealed these NEFAs interacted with multiple genes. KEGG pathway analysis on these genes suggested that differentially expressed NEFA may impact PPAR and cAMP signaling pathways.

LIMITATIONS

Inclusion of diverse populations for evaluation is warranted. Biomarkers identified in this study require samples that are more in line with the experimental design for external validation, and further basic research is necessary to validate the potential depressive mechanisms of NEFA.

CONCLUSIONS

The overall reduction in NEFA expression in first-episode adolescent depression patients suggests a potential mediation of depression symptoms through cAMP and PPAR signaling pathways. NEFA levels show promise as a diagnostic tool for identifying first-episode adolescent depression patients.

Individualized prediction of anxiety and depressive symptoms using gray matter volume in a non-clinical population

Machine learning is an emerging tool in clinical psychology and neuroscience for the individualized prediction of psychiatric symptoms. However, its application in non-clinical populations is still in its infancy. Given the widespread morphological changes observed in psychiatric disorders, our study applies five supervised machine learning regression algorithms-ridge regression, support vector regression, partial least squares regression, least absolute shrinkage and selection operator regression, and Elastic-Net regression-to predict anxiety and depressive symptom scores. We base these predictions on the whole-brain gray matter volume in a large non-clinical sample (n = 425). Our results demonstrate that machine learning algorithms can effectively predict individual variability in anxiety and depressive symptoms, as measured by the Mood and Anxiety Symptoms Questionnaire. The most discriminative features contributing to the prediction models were primarily located in the prefrontal-parietal, temporal, visual, and sub-cortical regions (e.g. amygdala, hippocampus, and putamen). These regions showed distinct patterns for anxious arousal and high positive affect in three of the five models (partial least squares regression, support vector regression, and ridge regression). Importantly, these predictions were consistent across genders and robust to demographic variability (e.g. age, parental education, etc.). Our findings offer critical insights into the distinct brain morphological patterns underlying specific components of anxiety and depressive symptoms, supporting the existing tripartite theory from a neuroimaging perspective.

Prediction of Early Antidepressant Efficacy in Patients with Major Depressive Disorder Based on Multidimensional Features of rs-fMRI and P11 Gene DNA Methylation: Prédiction de l'efficacité précoce d'un antidépresseur chez des patients souffrant du trouble dépressif majeur d'après les caractéristiques multidimensionnelles de la méthylation de l'ADN du gène P11 et de la IRMf-rs

OBJECTIVE

This study established a machine learning model based on the multidimensional data of resting-state functional activity of the brain and P11 gene DNA methylation to predict the early efficacy of antidepressant treatment in patients with major depressive disorder (MDD).

METHODS

A total of 98 Han Chinese MDD were analysed in this study. Patients were divided into 51 responders and 47 nonresponders according to whether the Hamilton Depression Rating Scale-17 items (HAMD-17) reduction rate was ≥50% after 2 weeks of antidepressant treatment. At baseline, the Illumina HiSeq Platform was used to detect the methylation of 74 CpG sites of the P11 gene in peripheral blood samples. Resting-state functional magnetic resonance imaging (rs-fMRI) scan detected the amplitude of low-frequency fluctuations (ALFF), regional homogeneity (ReHo), and functional connectivity (FC) in 116 brain regions. The least absolute shrinkage and selection operator analysis method was used to perform feature reduction and feature selection. Four typical machine learning methods were used to establish support vector machine (SVM), random forest (RF), Naïve Bayes (NB), and logistic regression (LR) prediction models based on different combinations of functional activity of the brain, P11 gene DNA methylation and clinical/demographic features after screening.

RESULTS

The SVM model based on ALFF, ReHo, FC, P11 methylation, and clinical/demographic features showed the best performance, with 95.92% predictive accuracy and 0.9967 area under the receiver operating characteristic curve, which was better than RF, NB, and LR models.

CONCLUSION

The multidimensional data features combining rs-fMRI, DNA methylation, and clinical/demographic features can predict the early antidepressant efficacy in MDD.

Treatment Response Prediction in Major Depressive Disorder Using Multimodal MRI and Clinical Data: Secondary Analysis of a Randomized Clinical Trial

OBJECTIVE

Response to antidepressant treatment in major depressive disorder varies substantially between individuals, which lengthens the process of finding effective treatment. The authors sought to determine whether a multimodal machine learning approach could predict early sertraline response in patients with major depressive disorder. They assessed the predictive contribution of MR neuroimaging and clinical assessments at baseline and after 1 week of treatment.

METHODS

This was a preregistered secondary analysis of data from the Establishing Moderators and Biosignatures of Antidepressant Response in Clinical Care (EMBARC) study, a multisite double-blind, placebo-controlled randomized clinical trial that included 296 adult outpatients with unmedicated recurrent or chronic major depressive disorder. MR neuroimaging and clinical data were collected before and after 1 week of treatment. Performance in predicting response and remission, collected after 8 weeks, was quantified using balanced accuracy (bAcc) and area under the receiver operating characteristic curve (AUROC) scores.

RESULTS

A total of 229 patients were included in the analyses (mean age, 38 years [SD=13]; 66% female). Internal cross-validation performance in predicting response to sertraline (bAcc=68% [SD=10], AUROC=0.73 [SD=0.03]) was significantly better than chance. External cross-validation on data from placebo nonresponders (bAcc=62%, AUROC=0.66) and placebo nonresponders who were switched to sertraline (bAcc=65%, AUROC=0.68) resulted in differences that suggest specificity for sertraline treatment compared with placebo treatment. Finally, multimodal models outperformed unimodal models.

CONCLUSIONS

The study results confirm that early sertraline treatment response can be predicted; that the models are sertraline specific compared with placebo; that prediction benefits from integrating multimodal MRI data with clinical data; and that perfusion imaging contributes most to these predictions. Using this approach, a lean and effective protocol could individualize sertraline treatment planning to improve psychiatric care.

Multimodal imaging measures in the prediction of clinical response to deep brain stimulation for refractory depression: A machine learning approach

OBJECTIVES

This study compared machine learning models using unimodal imaging measures and combined multi-modal imaging measures for deep brain stimulation (DBS) outcome prediction in treatment resistant depression (TRD).

METHODS

Regional brain glucose metabolism (CMRGlu), cerebral blood flow (CBF), and grey matter volume (GMV) were measured at baseline using 18F-fluorodeoxy glucose (18F-FDG) positron emission tomography (PET), arterial spin labelling (ASL) magnetic resonance imaging (MRI), and T1-weighted MRI, respectively, in 19 patients with TRD receiving subcallosal cingulate (SCC)-DBS. Responders (n = 9) were defined by a 50% reduction in HAMD-17 at 6 months from the baseline. Using an atlas-based approach, values of each measure were determined for pre-selected brain regions. OneR feature selection algorithm and the naïve Bayes model was used for classification. Leave-out-one cross validation was used for classifier evaluation.

RESULTS

The performance accuracy of the CMRGlu classification model (84%) was greater than CBF (74%) or GMV (74%) models. The classification model using the three image modalities together led to a similar accuracy (84%0 compared to the CMRGlu classification model.

CONCLUSIONS

CMRGlu imaging measures may be useful for the development of multivariate prediction models for SCC-DBS studies for TRD. The future of multivariate methods for multimodal imaging may rest on the selection of complementing features and the developing better models.Clinical Trial Registration: ClinicalTrials.gov (#NCT01983904).

Predicting Depression Risk in Patients With Cancer Using Multimodal Data: Algorithm Development Study

BACKGROUND

Patients with cancer starting systemic treatment programs, such as chemotherapy, often develop depression. A prediction model may assist physicians and health care workers in the early identification of these vulnerable patients.

OBJECTIVE

This study aimed to develop a prediction model for depression risk within the first month of cancer treatment.

METHODS

We included 16,159 patients diagnosed with cancer starting chemo- or radiotherapy treatment between 2008 and 2021. Machine learning models (eg, least absolute shrinkage and selection operator [LASSO] logistic regression) and natural language processing models (Bidirectional Encoder Representations from Transformers [BERT]) were used to develop multimodal prediction models using both electronic health record data and unstructured text (patient emails and clinician notes). Model performance was assessed in an independent test set (n=5387, 33%) using area under the receiver operating characteristic curve (AUROC), calibration curves, and decision curve analysis to assess initial clinical impact use.

RESULTS

Among 16,159 patients, 437 (2.7%) received a depression diagnosis within the first month of treatment. The LASSO logistic regression models based on the structured data (AUROC 0.74, 95% CI 0.71-0.78) and structured data with email classification scores (AUROC 0.74, 95% CI 0.71-0.78) had the best discriminative performance. The BERT models based on clinician notes and structured data with email classification scores had AUROCs around 0.71. The logistic regression model based on email classification scores alone performed poorly (AUROC 0.54, 95% CI 0.52-0.56), and the model based solely on clinician notes had the worst performance (AUROC 0.50, 95% CI 0.49-0.52). Calibration was good for the logistic regression models, whereas the BERT models produced overly extreme risk estimates even after recalibration. There was a small range of decision thresholds for which the best-performing model showed promising clinical effectiveness use. The risks were underestimated for female and Black patients.

CONCLUSIONS

The results demonstrated the potential and limitations of machine learning and multimodal models for predicting depression risk in patients with cancer. Future research is needed to further validate these models, refine the outcome label and predictors related to mental health, and address biases across subgroups.

Non-linear processing and reinforcement learning to predict rTMS treatment response in depression

BACKGROUND

Forecasting the efficacy of repetitive transcranial magnetic stimulation (rTMS) therapy can lead to substantial time and cost savings by preventing futile treatments. To achieve this objective, we've formulated a machine learning approach aimed at categorizing patients with major depressive disorder (MDD) into two groups: individuals who respond (R) positively to rTMS treatment and those who do not respond (NR).

METHODS

Preceding the commencement of treatment, we obtained resting-state EEG data from 106 patients diagnosed with MDD, employing 32 electrodes for data collection. These patients then underwent a 7-week course of rTMS therapy, and 54 of them exhibited positive responses to the treatment. Employing Independent Component Analysis (ICA) on the EEG data, we successfully pinpointed relevant brain sources that could potentially serve as markers of neural activity within the dorsolateral prefrontal cortex (DLPFC). These identified sources were further scrutinized to estimate the sources of activity within the sensor domain. Then, we integrated supplementary physiological data and implemented specific criteria to yield more realistic estimations when compared to conventional EEG analysis. In the end, we selected components corresponding to the DLPFC region within the sensor domain. Features were derived from the time-series data of these relevant independent components. To identify the most significant features, we used Reinforcement Learning (RL). In categorizing patients into two groups - R and NR to rTMS treatment - we utilized three distinct classification algorithms including K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Multilayer Perceptron (MLP). We assessed the performance of these classifiers through a ten-fold cross-validation method. Additionally, we conducted a statistical test to evaluate the discriminative capacity of these features between responders and non-responders, opening the door for further exploration in this field.

RESULTS

We identified EEG features that can anticipate the response to rTMS treatment. The most robust discriminators included EEG beta power, the sum of bispectrum diagonal elements in the delta and beta frequency bands. When these features were combined into a single vector, the classification of responders and non-responders achieved impressive performance, with an accuracy of 95.28 %, specificity at 94.23 %, sensitivity reaching 96.29 %, and precision standing at 94.54 %, all achieved using SVM.

CONCLUSIONS

The results of this study suggest that the proposed approach, utilizing power, non-linear, and bispectral features extracted from relevant independent component time-series, has the capability to forecast the treatment outcome of rTMS for MDD patients based solely on a single pre-treatment EEG recording session. The achieved findings demonstrate the superior performance of our method compared to previous techniques.

Brain multi-contrast, multi-atlas segmentation of diffusion tensor imaging and ensemble learning automatically diagnose late-life depression

We investigated the potential of machine learning for diagnostic classification in late-life major depression based on an advanced whole brain white matter segmentation framework. Twenty-six late-life depression and 12 never depressed individuals aged > 55 years, matched for age, MMSE, and education underwent brain diffusion tensor imaging and a multi-contrast, multi-atlas segmentation in MRIcloud. Fractional anisotropy volume, mean fractional anisotropy, trace, axial and radial diffusivity (RD) extracted from 146 white matter parcels for each subject were used to train and test the AdaBoost classifier using stratified 12-fold cross validation. Performance was evaluated using various measures. The statistical power of the classifier was assessed using label permutation test. Statistical analysis did not yield significant differences in DTI measures between the groups. The classifier achieved a balanced accuracy of 71% and an Area Under the Receiver Operator Characteristic Curve (ROC-AUC) of 0.81 by trace, and a balanced accuracy of 70% and a ROC-AUC of 0.80 by RD, in limbic, cortico-basal ganglia-thalamo-cortical loop, brainstem, external and internal capsules, callosal and cerebellar structures. Both indices shared important structures for classification, while fornix was the most important structure for classification by both indices. The classifier proved statistically significant, as trace and RD ROC-AUC scores after permutation were lower than those obtained with the actual data (P = 0.022 and P = 0.024, respectively). Similar results were obtained with the Gradient Boosting classifier, whereas the RBF-kernel Support Vector Machine with k-best feature selection did not exceed the chance level. Finally, AdaBoost significantly predicted the class using all features together. Limitations are discussed. The results encourage further investigation of the implemented methods for computer aided diagnostics and anatomically informed therapeutics.

Machine Learning Classification of Psychiatric Data Associated with Compensation Claims for Patient Injuries

BACKGROUND

Adverse events are common in health care. In psychiatric treatment, compensation claims for patient injuries appear to be less common than in other medical specialties. The most common types of patient injury claims in psychiatry include diagnostic flaws, unprevented suicide, or coercive treatment deemed as unnecessary or harmful.

OBJECTIVES

The objective was to study whether it is possible to form different categories of patient injury types associated with the psychiatric evaluations of compensation claims and to base machine learning classification on these categories. Further, the binary classification of positive and negative decisions for compensation claims was the other objective.

METHODS

Finnish psychiatric specialist evaluations for the compensation claims of patient injuries were classified into six different categories called classes applying the machine learning methods of artificial intelligence. In addition, another classification of the same data into two classes was performed to test whether it was possible to classify data cases according to their known decisions, either accepted or declined compensation claim.

RESULTS

The former classification task produced relatively good classification results subject to separating between different classes. Instead, the latter was more complex. However, classification accuracies of both tasks could be improved by using the generation of artificial data cases in the preprocessing phase before classifications. This preprocessing improved the classification accuracy of six classes up to 88% when the method of random forests was used for classification and that of the binary classification to 89%.

CONCLUSION

The results show that the objectives defined were possible to solve reasonably.

The diagnostic performance of machine learning based on resting-state functional magnetic resonance imaging data for major depressive disorders: a systematic review and meta-analysis

Objective

Machine learning (ML) has been widely used to detect and evaluate major depressive disorder (MDD) using neuroimaging data, i.e., resting-state functional magnetic resonance imaging (rs-fMRI). However, the diagnostic efficiency is unknown. The aim of the study is to conduct an updated meta-analysis to evaluate the diagnostic performance of ML based on rs-fMRI data for MDD.

Methods

English databases were searched for relevant studies. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) was used to assess the methodological quality of the included studies. A random-effects meta-analytic model was implemented to investigate the diagnostic efficiency, including sensitivity, specificity, diagnostic odds ratio (DOR), and area under the curve (AUC). Regression meta-analysis and subgroup analysis were performed to investigate the cause of heterogeneity.

Results

Thirty-one studies were included in this meta-analysis. The pooled sensitivity, specificity, DOR, and AUC with 95% confidence intervals were 0.80 (0.75, 0.83), 0.83 (0.74, 0.82), 14.00 (9, 22.00), and 0.86 (0.83, 0.89), respectively. Substantial heterogeneity was observed among the studies included. The meta-regression showed that the leave-one-out cross-validation (loocv) (sensitivity: p < 0.01, specificity: p < 0.001), graph theory (sensitivity: p < 0.05, specificity: p < 0.01), n > 100 (sensitivity: p < 0.001, specificity: p < 0.001), simens equipment (sensitivity: p < 0.01, specificity: p < 0.001), 3.0T field strength (Sensitivity: p < 0.001, specificity: p = 0.04), and Beck Depression Inventory (BDI) (sensitivity: p = 0.04, specificity: p = 0.06) might be the sources of heterogeneity. Furthermore, the subgroup analysis showed that the sample size (n > 100: sensitivity: 0.71, specificity: 0.72, n < 100: sensitivity: 0.81, specificity: 0.79), the different levels of disease evaluated by the Hamilton Depression Rating Scale (HDRS/HAMD) (mild vs. moderate vs. severe: sensitivity: 0.52 vs. 0.86 vs. 0.89, specificity: 0.62 vs. 0.78 vs. 0.82, respectively), the depression scales in patients with comparable levels of severity. (BDI vs. HDRS/HAMD: sensitivity: 0.86 vs. 0.87, specificity: 0.78 vs. 0.80, respectively), and the features (graph vs. functional connectivity: sensitivity: 0.84 vs. 0.86, specificity: 0.76 vs. 0.78, respectively) selected might be the causes of heterogeneity.

Conclusion

ML showed high accuracy for the automatic diagnosis of MDD. Future studies are warranted to promote the potential use of these classification algorithms in clinical settings.

A Call to Action on Assessing and Mitigating Bias in Artificial Intelligence Applications for Mental Health

Advances in computer science and data-analytic methods are driving a new era in mental health research and application. Artificial intelligence (AI) technologies hold the potential to enhance the assessment, diagnosis, and treatment of people experiencing mental health problems and to increase the reach and impact of mental health care. However, AI applications will not mitigate mental health disparities if they are built from historical data that reflect underlying social biases and inequities. AI models biased against sensitive classes could reinforce and even perpetuate existing inequities if these models create legacies that differentially impact who is diagnosed and treated, and how effectively. The current article reviews the health-equity implications of applying AI to mental health problems, outlines state-of-the-art methods for assessing and mitigating algorithmic bias, and presents a call to action to guide the development of fair-aware AI in psychological science.

Brain morphometric features predict depression symptom phenotypes in late-life depression using a deep learning model

Objectives

Our objective was to use deep learning models to identify underlying brain regions associated with depression symptom phenotypes in late-life depression (LLD).

Participants

Diagnosed with LLD (N = 116) and enrolled in a prospective treatment study.

Design

Cross-sectional.

Measurements

Structural magnetic resonance imaging (sMRI) was used to predict five depression symptom phenotypes from the Hamilton and MADRS depression scales previously derived from factor analysis: (1) Anhedonia, (2) Suicidality, (3) Appetite, (4) Sleep Disturbance, and (5) Anxiety. Our deep learning model was deployed to predict each factor score via learning deep feature representations from 3D sMRI patches in 34 a priori regions-of-interests (ROIs). ROI-level prediction accuracy was used to identify the most discriminative brain regions associated with prediction of factor scores representing each of the five symptom phenotypes.

Results

Factor-level results found significant predictive models for Anxiety and Suicidality factors. ROI-level results suggest the most LLD-associated discriminative regions in predicting all five symptom factors were located in the anterior cingulate and orbital frontal cortex.

Conclusions

We validated the effectiveness of using deep learning approaches on sMRI for predicting depression symptom phenotypes in LLD. We were able to identify deep embedded local morphological differences in symptom phenotypes in the brains of those with LLD, which is promising for symptom-targeted treatment of LLD. Future research with machine learning models integrating multimodal imaging and clinical data can provide additional discriminative information.

Sampling inequalities affect generalization of neuroimaging-based diagnostic classifiers in psychiatry

BACKGROUND

The development of machine learning models for aiding in the diagnosis of mental disorder is recognized as a significant breakthrough in the field of psychiatry. However, clinical practice of such models remains a challenge, with poor generalizability being a major limitation.

METHODS

Here, we conducted a pre-registered meta-research assessment on neuroimaging-based models in the psychiatric literature, quantitatively examining global and regional sampling issues over recent decades, from a view that has been relatively underexplored. A total of 476 studies (n = 118,137) were included in the current assessment. Based on these findings, we built a comprehensive 5-star rating system to quantitatively evaluate the quality of existing machine learning models for psychiatric diagnoses.

RESULTS

A global sampling inequality in these models was revealed quantitatively (sampling Gini coefficient (G) = 0.81, p < .01), varying across different countries (regions) (e.g., China, G = 0.47; the USA, G = 0.58; Germany, G = 0.78; the UK, G = 0.87). Furthermore, the severity of this sampling inequality was significantly predicted by national economic levels (β =  - 2.75, p < .001, R²_adj = 0.40; r =  - .84, 95% CI: - .41 to - .97), and was plausibly predictable for model performance, with higher sampling inequality for reporting higher classification accuracy. Further analyses showed that lack of independent testing (84.24% of models, 95% CI: 81.0-87.5%), improper cross-validation (51.68% of models, 95% CI: 47.2-56.2%), and poor technical transparency (87.8% of models, 95% CI: 84.9-90.8%)/availability (80.88% of models, 95% CI: 77.3-84.4%) are prevailing in current diagnostic classifiers despite improvements over time. Relating to these observations, model performances were found decreased in studies with independent cross-country sampling validations (all p < .001, BF₁₀ > 15). In light of this, we proposed a purpose-built quantitative assessment checklist, which demonstrated that the overall ratings of these models increased by publication year but were negatively associated with model performance.

CONCLUSIONS

Together, improving sampling economic equality and hence the quality of machine learning models may be a crucial facet to plausibly translating neuroimaging-based diagnostic classifiers into clinical practice.

Deep learning and machine learning in psychiatry: a survey of current progress in depression detection, diagnosis and treatment

Informatics paradigms for brain and mental health research have seen significant advances in recent years. These developments can largely be attributed to the emergence of new technologies such as machine learning, deep learning, and artificial intelligence. Data-driven methods have the potential to support mental health care by providing more precise and personalised approaches to detection, diagnosis, and treatment of depression. In particular, precision psychiatry is an emerging field that utilises advanced computational techniques to achieve a more individualised approach to mental health care. This survey provides an overview of the ways in which artificial intelligence is currently being used to support precision psychiatry. Advanced algorithms are being used to support all phases of the treatment cycle. These systems have the potential to identify individuals suffering from mental health conditions, allowing them to receive the care they need and tailor treatments to individual patients who are mostly to benefit. Additionally, unsupervised learning techniques are breaking down existing discrete diagnostic categories and highlighting the vast disease heterogeneity observed within depression diagnoses. Artificial intelligence also provides the opportunity to shift towards evidence-based treatment prescription, moving away from existing methods based on group averages. However, our analysis suggests there are several limitations currently inhibiting the progress of data-driven paradigms in care. Significantly, none of the surveyed articles demonstrate empirically improved patient outcomes over existing methods. Furthermore, greater consideration needs to be given to uncertainty quantification, model validation, constructing interdisciplinary teams of researchers, improved access to diverse data and standardised definitions within the field. Empirical validation of computer algorithms via randomised control trials which demonstrate measurable improvement to patient outcomes are the next step in progressing models to clinical implementation.

Resting-state functional connectivity disruption between the left and right pallidum as a biomarker for subthreshold depression

Although the identification of late adolescents with subthreshold depression (StD) may provide a basis for developing effective interventions that could lead to a reduction in the prevalence of StD and prevent the development of major depressive disorder, knowledge about the neural basis of StD remains limited. The purpose of this study was to develop a generalizable classifier for StD and to shed light on the underlying neural mechanisms of StD in late adolescents. Resting-state functional magnetic resonance imaging data of 91 individuals (30 StD subjects, 61 healthy controls) were included to build an StD classifier, and eight functional connections were selected by using the combination of two machine learning algorithms. We applied this biomarker to an independent cohort (n = 43) and confirmed that it showed generalization performance (area under the curve = 0.84/0.75 for the training/test datasets). Moreover, the most important functional connection was between the left and right pallidum, which may be related to clinically important dysfunctions in subjects with StD such as anhedonia and hyposensitivity to rewards. Investigation of whether modulation of the identified functional connections can be an effective treatment for StD may be an important topic of future research.

Machine learning approaches and non-linear processing of extracted components in frontal region to predict rTMS treatment response in major depressive disorder

Predicting the therapeutic result of repetitive transcranial magnetic stimulation (rTMS) treatment could save time and costs as ineffective treatment can be avoided. To this end, we presented a machine-learning-based strategy for classifying patients with major depression disorder (MDD) into responders (R) and nonresponders (NR) to rTMS treatment. Resting state EEG data were recorded using 32 electrodes from 88 MDD patients before treatment. Then, patients underwent 7 weeks of rTMS, and 46 of them responded to treatment. By applying Independent Component Analysis (ICA) on EEG, we identified the relevant brain sources as possible indicators of neural activity in the dorsolateral prefrontal cortex (DLPFC). This was served through estimating the generators of activity in the sensor domain. Subsequently, we added physiological information and placed certain terms and conditions to offer a far more realistic estimation than the classic EEG. Ultimately, those components mapped in accordance with the region of the DLPFC in the sensor domain were chosen. Features extracted from the relevant ICs time series included permutation entropy (PE), fractal dimension (FD), Lempel-Ziv Complexity (LZC), power spectral density, correlation dimension (CD), features based on bispectrum, frontal and prefrontal cordance, and a combination of them. The most relevant features were selected by a Genetic Algorithm (GA). For classifying two groups of R and NR, K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Multilayer Perceptron (MLP) were applied to predict rTMS treatment response. To evaluate the performance of classifiers, a 10-fold cross-validation method was employed. A statistical test was used to assess the capability of features in differentiating R and NR for further research. EEG characteristics that can predict rTMS treatment response were discovered. The strongest discriminative indicators were EEG beta power, the sum of bispectrum diagonal elements in delta and beta bands, and CD. The Combined feature vector classified R and NR with a high performance of 94.31% accuracy, 92.85% specificity, 95.65% sensitivity, and 92.85% precision using SVM. This result indicates that our proposed method with power and nonlinear and bispectral features from relevant ICs time-series can predict the treatment outcome of rTMS for MDD patients only by one session pretreatment EEG recording. The obtained results show that the proposed method outperforms previous methods.

Automatic diagnosis of late-life depression by 3D convolutional neural networks and cross-sample Entropy analysis from resting-state fMRI

Resting-state fMRI has been widely used in investigating the pathophysiology of late-life depression (LLD). Unlike the conventional linear approach, cross-sample entropy (CSE) analysis shows the nonlinear property in fMRI signals between brain regions. Moreover, recent advances in deep learning, such as convolutional neural networks (CNNs), provide a timely application for understanding LLD. Accurate and prompt diagnosis is essential in LLD; hence, this study aimed to combine CNN and CSE analysis to discriminate LLD patients and non-depressed comparison older adults based on brain resting-state fMRI signals. Seventy-seven older adults, including 49 patients and 28 comparison older adults, were included for fMRI scans. Three-dimensional CSEs with volumes corresponding to 90 seed regions of interest of each participant were developed and fed into models for disease classification and depression severity prediction. We obtained a diagnostic accuracy > 85% in the superior frontal gyrus (left dorsolateral and right orbital parts), left insula, and right middle occipital gyrus. With a mean root-mean-square error (RMSE) of 2.41, three separate models were required to predict depressive symptoms in the severe, moderate, and mild depression groups. The CSE volumes in the left inferior parietal lobule, left parahippocampal gyrus, and left postcentral gyrus performed best in each respective model. Combined complexity analysis and deep learning algorithms can classify patients with LLD from comparison older adults and predict symptom severity based on fMRI data. Such application can be utilized in precision medicine for disease detection and symptom monitoring in LLD.

Measuring depression severity based on facial expression and body movement using deep convolutional neural network

Introduction

Real-time evaluations of the severity of depressive symptoms are of great significance for the diagnosis and treatment of patients with major depressive disorder (MDD). In clinical practice, the evaluation approaches are mainly based on psychological scales and doctor-patient interviews, which are time-consuming and labor-intensive. Also, the accuracy of results mainly depends on the subjective judgment of the clinician. With the development of artificial intelligence (AI) technology, more and more machine learning methods are used to diagnose depression by appearance characteristics. Most of the previous research focused on the study of single-modal data; however, in recent years, many studies have shown that multi-modal data has better prediction performance than single-modal data. This study aimed to develop a measurement of depression severity from expression and action features and to assess its validity among the patients with MDD.

Methods

We proposed a multi-modal deep convolutional neural network (CNN) to evaluate the severity of depressive symptoms in real-time, which was based on the detection of patients' facial expression and body movement from videos captured by ordinary cameras. We established behavioral depression degree (BDD) metrics, which combines expression entropy and action entropy to measure the depression severity of MDD patients.

Results

We found that the information extracted from different modes, when integrated in appropriate proportions, can significantly improve the accuracy of the evaluation, which has not been reported in previous studies. This method presented an over 74% Pearson similarity between BDD and self-rating depression scale (SDS), self-rating anxiety scale (SAS), and Hamilton depression scale (HAMD). In addition, we tracked and evaluated the changes of BDD in patients at different stages of a course of treatment and the results obtained were in agreement with the evaluation from the scales.

Discussion

The BDD can effectively measure the current state of patients' depression and its changing trend according to the patient's expression and action features. Our model may provide an automatic auxiliary tool for the diagnosis and treatment of MDD.

Computational approaches to treatment response prediction in major depression using brain activity and behavioral data: A systematic review

Major depressive disorder is a heterogeneous diagnostic category with multiple available treatments. With the goal of optimizing treatment selection, researchers are developing computational models that attempt to predict treatment response based on various pretreatment measures. In this paper, we review studies that use brain activity data to predict treatment response. Our aim is to highlight and clarify important methodological differences between various studies that relate to the incorporation of domain knowledge, specifically within two approaches delineated as data-driven and theory-driven. We argue that theory-driven generative modeling, which explicitly models information processing in the brain and thus can capture disease mechanisms, is a promising emerging approach that is only beginning to be utilized in treatment response prediction. The predictors extracted via such models could improve interpretability, which is critical for clinical decision-making. We also identify several methodological limitations across the reviewed studies and provide suggestions for addressing them. Namely, we consider problems with dichotomizing treatment outcomes, the importance of investigating more than one treatment in a given study for differential treatment response predictions, the need for a patient-centered approach for defining treatment outcomes, and finally, the use of internal and external validation methods for improving model generalizability.

Integrated multimodal artificial intelligence framework for healthcare applications

Artificial intelligence (AI) systems hold great promise to improve healthcare over the next decades. Specifically, AI systems leveraging multiple data sources and input modalities are poised to become a viable method to deliver more accurate results and deployable pipelines across a wide range of applications. In this work, we propose and evaluate a unified Holistic AI in Medicine (HAIM) framework to facilitate the generation and testing of AI systems that leverage multimodal inputs. Our approach uses generalizable data pre-processing and machine learning modeling stages that can be readily adapted for research and deployment in healthcare environments. We evaluate our HAIM framework by training and characterizing 14,324 independent models based on HAIM-MIMIC-MM, a multimodal clinical database (N = 34,537 samples) containing 7279 unique hospitalizations and 6485 patients, spanning all possible input combinations of 4 data modalities (i.e., tabular, time-series, text, and images), 11 unique data sources and 12 predictive tasks. We show that this framework can consistently and robustly produce models that outperform similar single-source approaches across various healthcare demonstrations (by 6–33%), including 10 distinct chest pathology diagnoses, along with length-of-stay and 48 h mortality predictions. We also quantify the contribution of each modality and data source using Shapley values, which demonstrates the heterogeneity in data modality importance and the necessity of multimodal inputs across different healthcare-relevant tasks. The generalizable properties and flexibility of our Holistic AI in Medicine (HAIM) framework could offer a promising pathway for future multimodal predictive systems in clinical and operational healthcare settings.

Prediction of remission among patients with a major depressive disorder based on the resting-state functional connectivity of emotion regulation networks

The prediction of antidepressant response is critical for psychiatrists to select the initial antidepressant drug for patients with major depressive disorders (MDD). The implicated brain networks supporting emotion regulation (ER) are critical in the pathophysiology of MDD and the prediction of antidepressant response. Therefore, the primary aim of the current study was to identify the neuroimaging biomarkers for the prediction of remission in patients with MDD based on the resting-state functional connectivity (rsFC) of the ER networks. A total of 81 unmedicated adult MDD patients were investigated and they underwent resting-state functional magnetic resonance imagining (fMRI) scans. The patients were treated with escitalopram for 12 weeks. The 17-item Hamilton depression rating scale was used for assessing remission. The 36 seed regions from predefined ER networks were selected and the rsFC matrix was caculated for each participant. The support vector machine algorithm was employed to construct prediction model, which separated the patients with remission from those with non-remission. And leave-one-out cross-validation and the area under the curve (AUC) of the receiver operating characteristic were used for evaluating the performance of the model. The accuracy of the prediction model was 82.08% (sensitivity = 71.43%, specificity = 89.74%, AUC = 0.86). The rsFC between the left medial superior frontal gyrus and the right inferior frontal gyrus as well as the precuneus were the features with the highest discrimination ability in predicting remission from escitalopram among the MDD patients. Results from our study demonstrated that rsFC of the ER brain networks are potential predictors for the response of antidepressant drugs. The trial name: appropriate technology study of MDD diagnosis and treatment based on objective indicators and measurement. URL: http://www.chictr.org.cn/showproj.aspx?proj=21377 . Registration number: ChiCTR-OOC-17012566.

Associations of depression and regional brain structure across the adult lifespan: Pooled analyses of six population-based and two clinical cohort studies in the European Lifebrain consortium

OBJECTIVE

Major depressive disorder has been associated with lower prefrontal thickness and hippocampal volume, but it is unknown whether this association also holds for depressive symptoms in the general population. We investigated associations of depressive symptoms and depression status with brain structures across population-based and patient-control cohorts, and explored whether these associations are similar over the lifespan and across sexes.

METHODS

We included 3,447 participants aged 18-89 years from six population-based and two clinical patient-control cohorts of the European Lifebrain consortium. Cross-sectional meta-analyses using individual person data were performed for associations of depressive symptoms and depression status with FreeSurfer-derived thickness of bilateral rostral anterior cingulate cortex (rACC) and medial orbitofrontal cortex (mOFC), and hippocampal and total grey matter volume (GMV), separately for population-based and clinical cohorts.

RESULTS

Across patient-control cohorts, depressive symptoms and presence of mild-to-severe depression were associated with lower mOFC thickness (r_symptoms = -0.15/ r_status = -0.22), rACC thickness (r_symptoms = -0.20/ r_status = -0.25), hippocampal volume (r_symptoms = -0.13/ r_status = 0.13) and total GMV (r_symptoms = -0.21/ r_status = -0.25). Effect sizes were slightly larger for presence of moderate-to-severe depression. Associations were similar across age groups and sex. Across population-based cohorts, no associations between depression and brain structures were observed.

CONCLUSIONS

Fitting with previous meta-analyses, depressive symptoms and depression status were associated with lower mOFC, rACC thickness, and hippocampal and total grey matter volume in clinical patient-control cohorts, although effect sizes were small. The absence of consistent associations in population-based cohorts with mostly mild depressive symptoms, suggests that significantly lower thickness and volume of the studied brain structures are only detectable in clinical populations with more severe depressive symptoms.

Resting-state neural signal variability in women with depressive disorders

Aberrant activity and connectivity in default mode (DMN), frontoparietal (FPN), and salience (SN) network regions is well-documented in depression. Recent neuroimaging research suggests that altered variability in the blood oxygen level-dependent (BOLD) signal may disrupt normal network integration and be an important novel predictor of psychopathology. However, no studies have yet determined the relationship between resting-state BOLD signal variability and depressive disorders nor applied BOLD signal variability features to the classification of depression history using machine learning (ML). We collected resting-state fMRI data for 79 women with different depression histories: no history, past history, and current depressive disorder. We tested voxelwise differences in BOLD signal variability related to depression group and severity. We also investigated whether BOLD signal variability of DMN, FPN, and SN regions could predict depression history group using a supervised random forest ML model. Results indicated that individuals with any history of depression had significantly decreased BOLD signal variability in the left and right cerebellum and right parietal cortex (pFWE <0.05). Furthermore, greater depression severity was also associated with reduced BOLD signal variability in the cerebellum. A random forest model classified participant depression history with 74% accuracy, with the ventral anterior cingulate cortex of the DMN as the most important variable in the model. These findings provide novel support for resting-state BOLD signal variability as a marker of neural dysfunction in depression and implicate decreased neural signal variability in the pathophysiology of depression.

Personalized Diagnosis and Treatment for Neuroimaging in Depressive Disorders

Depressive disorders are highly heterogeneous in nature. Previous studies have not been useful for the clinical diagnosis and prediction of outcomes of major depressive disorder (MDD) at the individual level, although they provide many meaningful insights. To make inferences beyond group-level analyses, machine learning (ML) techniques can be used for the diagnosis of subtypes of MDD and the prediction of treatment responses. We searched PubMed for relevant studies published until December 2021 that included depressive disorders and applied ML algorithms in neuroimaging fields for depressive disorders. We divided these studies into two sections, namely diagnosis and treatment outcomes, for the application of prediction using ML. Structural and functional magnetic resonance imaging studies using ML algorithms were included. Thirty studies were summarized for the prediction of an MDD diagnosis. In addition, 19 studies on the prediction of treatment outcomes for MDD were reviewed. We summarized and discussed the results of previous studies. For future research results to be useful in clinical practice, ML enabling individual inferences is important. At the same time, there are important challenges to be addressed in the future.

Simultaneous electroencephalography-functional magnetic resonance imaging for assessment of human brain function

Electroencephalography (EEG) and functional Magnetic Resonance Imaging (MRI) have long been used as tools to examine brain activity. Since both methods are very sensitive to changes of synaptic activity, simultaneous recording of EEG and fMRI can provide both high temporal and spatial resolution. Therefore, the two modalities are now integrated into a hybrid tool, EEG-fMRI, which encapsulates the useful properties of the two. Among other benefits, EEG-fMRI can contribute to a better understanding of brain connectivity and networks. This review lays its focus on the methodologies applied in performing EEG-fMRI studies, namely techniques used for the recording of EEG inside the scanner, artifact removal, and statistical analysis of the fMRI signal. We will investigate simultaneous resting-state and task-based EEG-fMRI studies and discuss their clinical and technological perspectives. Moreover, it is established that the brain regions affected by a task-based neural activity might not be limited to the regions in which they have been initiated. Advanced methods can help reveal the regions responsible for or affected by a developed neural network. Therefore, we have also looked into studies related to characterization of structure and dynamics of brain networks. The reviewed literature suggests that EEG-fMRI can provide valuable complementary information about brain neural networks and functions.

Multi-centre classification of functional neurological disorders based on resting-state functional connectivity

•

Using machine learning on multi-centre data, FND patients were successfully classified with an accuracy of 72%.

•

The angular- and supramarginal gyri, cingular- and insular cortex, and the hippocampus were the most discriminant regions.

•

To provide diagnostic utility, future studies must include patients with similar symptoms but different diagnoses.

Background

Patients suffering from functional neurological disorder (FND) experience disabling neurological symptoms not caused by an underlying classical neurological disease (such as stroke or multiple sclerosis). The diagnosis is made based on reliable positive clinical signs, but clinicians often require additional time- and cost consuming medical tests and examinations. Resting-state functional connectivity (RS FC) showed its potential as an imaging-based adjunctive biomarker to help distinguish patients from healthy controls and could represent a “rule-in” procedure to assist in the diagnostic process. However, the use of RS FC depends on its applicability in a multi-centre setting, which is particularly susceptible to inter-scanner variability. The aim of this study was to test the robustness of a classification approach based on RS FC in a multi-centre setting.

Methods

This study aimed to distinguish 86 FND patients from 86 healthy controls acquired in four different centres using a multivariate machine learning approach based on whole-brain resting-state functional connectivity. First, previously published results were replicated in each centre individually (intra-centre cross-validation) and its robustness across inter-scanner variability was assessed by pooling all the data (pooled cross-validation). Second, we evaluated the generalizability of the method by using data from each centre once as a test set, and the data from the remaining centres as a training set (inter-centre cross-validation).

Results

FND patients were successfully distinguished from healthy controls in the replication step (accuracy of 74%) as well as in each individual additional centre (accuracies of 73%, 71% and 70%). The pooled cross validation confirmed that the classifier was robust with an accuracy of 72%. The results survived post-hoc adjustment for anxiety, depression, psychotropic medication intake, and symptom severity. The most discriminant features involved the angular- and supramarginal gyri, sensorimotor cortex, cingular- and insular cortex, and hippocampal regions. The inter-centre validation step did not exceed chance level (accuracy below 50%).

Conclusions

The results demonstrate the applicability of RS FC to correctly distinguish FND patients from healthy controls in different centres and its robustness against inter-scanner variability. In order to generalize its use across different centres and aim for clinical application, future studies should work towards optimization of acquisition parameters and include neurological and psychiatric control groups presenting with similar symptoms.

Application of machine learning methods in clinical trials for precision medicine

OBJECTIVE

A key component for precision medicine is a good prediction algorithm for patients' response to treatments. We aim to implement machine learning (ML) algorithms into the response-adaptive randomization (RAR) design and improve the treatment outcomes.

MATERIALS AND METHODS

We incorporated 9 ML algorithms to model the relationship of patient responses and biomarkers in clinical trial design. Such a model predicted the response rate of each treatment for each new patient and provide guidance for treatment assignment. Realizing that no single method may fit all trials well, we also built an ensemble of these 9 methods. We evaluated their performance through quantifying the benefits for trial participants, such as the overall response rate and the percentage of patients who receive their optimal treatments.

RESULTS

Simulation studies showed that the adoption of ML methods resulted in more personalized optimal treatment assignments and higher overall response rates among trial participants. Compared with each individual ML method, the ensemble approach achieved the highest response rate and assigned the largest percentage of patients to their optimal treatments. For the real-world study, we successfully showed the potential improvements if the proposed design had been implemented in the study.

CONCLUSION

In summary, the ML-based RAR design is a promising approach for assigning more patients to their personalized effective treatments, which makes the clinical trial more ethical and appealing. These features are especially desirable for late-stage cancer patients who have failed all the Food and Drug Administration (FDA)-approved treatment options and only can get new treatments through clinical trials.

Improving the diagnostic accuracy for major depressive disorder using machine learning algorithms integrating clinical and near-infrared spectroscopy data

BACKGROUND

Given that major depressive disorder (MDD) is both biologically and clinically heterogeneous, a diagnostic system integrating neurobiological markers and clinical characteristics would allow for better diagnostic accuracy and, consequently, treatment efficacy.

OBJECTIVE

Our study aimed to evaluate the discriminative and predictive ability of unimodal, bimodal, and multimodal approaches in a total of seven machine learning (ML) models-clinical, demographic, functional near-infrared spectroscopy (fNIRS), combinations of two unimodal models, as well as a combination of all three-for MDD.

METHODS

We recruited 65 adults with MDD and 68 matched healthy controls, who provided both sociodemographic and clinical information, and completed the HAM-D questionnaire. They were also subject to fNIRS measurement when participating in the verbal fluency task. Using the nested cross validation procedure, the classification performance of each ML model was evaluated based on the area under the receiver operating characteristic curve (ROC), balanced accuracy, sensitivity, and specificity.

RESULTS

The multimodal ML model was able to distinguish between depressed patients and healthy controls with the highest balanced accuracy of 87.98 ± 8.84% (AUC = 0.92; 95% CI (0.84-0.99) when compared with the uni- and bi-modal models.

CONCLUSIONS

Our multimodal ML model demonstrated the highest diagnostic accuracy for MDD. This reinforces the biological and clinical heterogeneity of MDD and highlights the potential of this model to improve MDD diagnosis rates. Furthermore, this model is cost-effective and clinically applicable enough to be established as a robust diagnostic system for MDD based on patients' biosignatures.

A neural network approach to optimising treatments for depression using data from specialist and community psychiatric services in Australia, New Zealand and Japan

This study investigated the application of a recurrent neural network for optimising pharmacological treatment for depression. A clinical dataset of 458 participants from specialist and community psychiatric services in Australia, New Zealand and Japan were extracted from an existing custom-built, web-based tool called Psynary . This data, which included baseline and self-completed reviews, was used to train and refine a novel algorithm which was a fully connected network feature extractor and long short-term memory algorithm was firstly trained in isolation and then integrated and annealed using slow learning rates due to the low dimensionality of the data. The accuracy of predicting depression remission before processing patient review data was 49.8%. After processing only 2 reviews, the accuracy was 76.5%. When considering a change in medication, the precision of changing medications was 97.4% and the recall was 71.4% . The medications with predicted best results were antipsychotics (88%) and selective serotonin reuptake inhibitors (87.9%). This is the first study that has created an all-in-one algorithm for optimising treatments for all subtypes of depression. Reducing treatment optimisation time for patients suffering with depression may lead to earlier remission and hence reduce the high levels of disability associated with the condition. Furthermore, in a setting where mental health conditions are increasing strain on mental health services, the utilisation of web-based tools for remote monitoring and machine/deep learning algorithms may assist clinicians in both specialist and primary care in extending specialist mental healthcare to a larger patient community.

Machine Learning in Healthcare

Recent advancements in Artificial Intelligence (AI) and Machine Learning (ML) technology have brought on substantial strides in predicting and identifying health emergencies, disease populations, and disease state and immune response, amongst a few. Although, skepticism remains regarding the practical application and interpretation of results from ML-based approaches in healthcare settings, the inclusion of these approaches is increasing at a rapid pace. Here we provide a brief overview of machine learning-based approaches and learning algorithms including supervised, unsupervised, and reinforcement learning along with examples. Second, we discuss the application of ML in several healthcare fields, including radiology, genetics, electronic health records, and neuroimaging. We also briefly discuss the risks and challenges of ML application to healthcare such as system privacy and ethical concerns and provide suggestions for future applications.

Machine learning in the prediction of depression treatment outcomes: a systematic review and meta-analysis

BACKGROUND

Multiple treatments are effective for major depressive disorder (MDD), but the outcomes of each treatment vary broadly among individuals. Accurate prediction of outcomes is needed to help select a treatment that is likely to work for a given person. We aim to examine the performance of machine learning methods in delivering replicable predictions of treatment outcomes.

METHODS

Of 7732 non-duplicate records identified through literature search, we retained 59 eligible reports and extracted data on sample, treatment, predictors, machine learning method, and treatment outcome prediction. A minimum sample size of 100 and an adequate validation method were used to identify adequate-quality studies. The effects of study features on prediction accuracy were tested with mixed-effects models. Fifty-four of the studies provided accuracy estimates or other estimates that allowed calculation of balanced accuracy of predicting outcomes of treatment.

RESULTS

Eight adequate-quality studies reported a mean accuracy of 0.63 [95% confidence interval (CI) 0.56-0.71], which was significantly lower than a mean accuracy of 0.75 (95% CI 0.72-0.78) in the other 46 studies. Among the adequate-quality studies, accuracies were higher when predicting treatment resistance (0.69) and lower when predicting remission (0.60) or response (0.56). The choice of machine learning method, feature selection, and the ratio of features to individuals were not associated with reported accuracy.

CONCLUSIONS

The negative relationship between study quality and prediction accuracy, combined with a lack of independent replication, invites caution when evaluating the potential of machine learning applications for personalizing the treatment of depression.

Machine Learning and Deep Learning for the Pharmacogenomics of Antidepressant Treatments

A growing body of evidence now proposes that machine learning and deep learning techniques can serve as a vital foundation for the pharmacogenomics of antidepressant treatments in patients with major depressive disorder (MDD). In this review, we focus on the latest developments for pharmacogenomics research using machine learning and deep learning approaches together with neuroimaging and multi-omics data. First, we review relevant pharmacogenomics studies that leverage numerous machine learning and deep learning techniques to determine treatment prediction and potential biomarkers for antidepressant treatments in MDD. In addition, we depict some neuroimaging pharmacogenomics studies that utilize various machine learning approaches to predict antidepressant treatment outcomes in MDD based on the integration of research on pharmacogenomics and neuroimaging. Moreover, we summarize the limitations in regard to the past pharmacogenomics studies of antidepressant treatments in MDD. Finally, we outline a discussion of challenges and directions for future research. In light of latest advancements in neuroimaging and multi-omics, various genomic variants and biomarkers associated with antidepressant treatments in MDD are being identified in pharmacogenomics research by employing machine learning and deep learning algorithms.

Autism Spectrum Disorder Studies Using fMRI Data and Machine Learning: A Review

Machine learning methods have been frequently applied in the field of cognitive neuroscience in the last decade. A great deal of attention has been attracted to introduce machine learning methods to study the autism spectrum disorder (ASD) in order to find out its neurophysiological underpinnings. In this paper, we presented a comprehensive review about the previous studies since 2011, which applied machine learning methods to analyze the functional magnetic resonance imaging (fMRI) data of autistic individuals and the typical controls (TCs). The all-round process was covered, including feature construction from raw fMRI data, feature selection methods, machine learning methods, factors for high classification accuracy, and critical conclusions. Applying different machine learning methods and fMRI data acquired from different sites, classification accuracies were obtained ranging from 48.3% up to 97%, and informative brain regions and networks were located. Through thorough analysis, high classification accuracies were found to usually occur in the studies which involved task-based fMRI data, single dataset for some selection principle, effective feature selection methods, or advanced machine learning methods. Advanced deep learning together with the multi-site Autism Brain Imaging Data Exchange (ABIDE) dataset became research trends especially in the recent 4 years. In the future, advanced feature selection and machine learning methods combined with multi-site dataset or easily operated task-based fMRI data may appear to have the potentiality to serve as a promising diagnostic tool for ASD.

Machine Learning-Based Definition of Symptom Clusters and Selection of Antidepressants for Depressive Syndrome

The current polythetic and operational criteria for major depression inevitably contribute to the heterogeneity of depressive syndromes. The heterogeneity of depressive syndrome has been criticized using the concept of language game in Wittgensteinian philosophy. Moreover, “a symptom- or endophenotype-based approach, rather than a diagnosis-based approach, has been proposed” as the “next-generation treatment for mental disorders” by Thomas Insel. Understanding the heterogeneity renders promise for personalized medicine to treat cases of depressive syndrome, in terms of both defining symptom clusters and selecting antidepressants. Machine learning algorithms have emerged as a tool for personalized medicine by handling clinical big data that can be used as predictors for subtype classification and treatment outcome prediction. The large clinical cohort data from the Sequenced Treatment Alternatives to Relieve Depression (STAR*D), Combining Medications to Enhance Depression Outcome (CO-MED), and the German Research Network on Depression (GRND) have recently began to be acknowledged as useful sources for machine learning-based depression research with regard to cost effectiveness and generalizability. In addition, noninvasive biological tools such as functional and resting state magnetic resonance imaging techniques are widely combined with machine learning methods to detect intrinsic endophenotypes of depression. This review highlights recent studies that have used clinical cohort or brain imaging data and have addressed machine learning-based approaches to defining symptom clusters and selecting antidepressants. Potentially applicable suggestions to realize machine learning-based personalized medicine for depressive syndrome are also provided herein.

Spectral clustering based on structural magnetic resonance imaging and its relationship with major depressive disorder and cognitive ability

There is increasing interest in using data-driven unsupervised methods to identify structural underpinnings of common mental illnesses, including major depressive disorder (MDD) and associated traits such as cognition. However, studies are often limited to severe clinical cases with small sample sizes and most do not include replication. Here, we examine two relatively large samples with structural magnetic resonance imaging (MRI), measures of lifetime MDD and cognitive variables: Generation Scotland (GS subsample, N = 980) and UK Biobank (UKB, N = 8,900), for discovery and replication, using an exploratory approach. Regional measures of FreeSurfer derived cortical thickness (CT), cortical surface area (CSA), cortical volume (CV) and subcortical volume (subCV) were input into a clustering process, controlling for common covariates. The main analysis steps involved constructing participant K-nearest neighbour graphs and graph partitioning with Markov stability to determine optimal clustering of participants. Resultant clusters were (1) checked whether they were replicated in an independent cohort and (2) tested for associations with depression status and cognitive measures. Participants separated into two clusters based on structural brain measurements in GS subsample, with large Cohen's d effect sizes between clusters in higher order cortical regions, commonly associated with executive function and decision making. Clustering was replicated in the UKB sample, with high correlations of cluster effect sizes for CT, CSA, CV and subCV between cohorts across regions. The identified clusters were not significantly different with respect to MDD case-control status in either cohort (GS subsample: p_FDR  = .2239-.6585; UKB: p_FDR  = .2003-.7690). Significant differences in general cognitive ability were, however, found between the clusters for both datasets, for CSA, CV and subCV (GS subsample: d = 0.2529-.3490, p_FDR  < .005; UKB: d = 0.0868-0.1070, p_FDR  < .005). Our results suggest that there are replicable natural groupings of participants based on cortical and subcortical brain measures, which may be related to differences in cognitive performance, but not to the MDD case-control status.

Systematic misestimation of machine learning performance in neuroimaging studies of depression

We currently observe a disconcerting phenomenon in machine learning studies in psychiatry: While we would expect larger samples to yield better results due to the availability of more data, larger machine learning studies consistently show much weaker performance than the numerous small-scale studies. Here, we systematically investigated this effect focusing on one of the most heavily studied questions in the field, namely the classification of patients suffering from Major Depressive Disorder (MDD) and healthy controls based on neuroimaging data. Drawing upon structural MRI data from a balanced sample of N = 1868 MDD patients and healthy controls from our recent international Predictive Analytics Competition (PAC), we first trained and tested a classification model on the full dataset which yielded an accuracy of 61%. Next, we mimicked the process by which researchers would draw samples of various sizes (N = 4 to N = 150) from the population and showed a strong risk of misestimation. Specifically, for small sample sizes (N = 20), we observe accuracies of up to 95%. For medium sample sizes (N = 100) accuracies up to 75% were found. Importantly, further investigation showed that sufficiently large test sets effectively protect against performance misestimation whereas larger datasets per se do not. While these results question the validity of a substantial part of the current literature, we outline the relatively low-cost remedy of larger test sets, which is readily available in most cases.

Prediction of repeated-dose intravenous ketamine response in major depressive disorder using the GWAS-based machine learning approach

Major depressive disorder (MDD) is one of the most common psychiatric disorders. Various clinical studies have shown that the N-methyl-D-aspartate (NMDA) receptor antagonist ketamine has rapid, robust, and sustained antidepressant effects. However, given the concerns about the adverse effects of ketamine on patients, it would be important to identify a set of biomarkers that could be used to predict clinical outcomes for its treatment. A total of 83 MDD patients received treatment with six ketamine infusions for up to 2 weeks and were classified into "responders" or "non-responders" based on an average change in the HAMD score >50% from baseline. A nested cross-validation approach was applied to prevent information leakage and overestimation of model performance. The initial dataset was divided randomly into training and test sets in a nested six-fold cross-validation. We first performed genome-wide logistic regression to find potentially significant variants related to treatment response and then selected the top SNPs based on the genetic association results using the random forests algorithm. Subsequently, six machine learning models were employed to construct prediction models by using ten-fold cross-validation. A series of model comparisons showed that the best performing fold was characterized by accuracy of 0.85, precision of 0.75, and a sensitivity of 1.00 with the support vector machine algorithm. Together, these findings demonstrated that the machine learning approach can predict the treatment outcomes of multiple ketamine infusions on the basis of the genotyping information of each participant.

Psychiatric Neural Networks and Precision Therapeutics by Machine Learning

Learning and environmental adaptation increase the likelihood of survival and improve the quality of life. However, it is often difficult to judge optimal behaviors in real life due to highly complex social dynamics and environment. Consequentially, many different brain regions and neuronal circuits are involved in decision-making. Many neurobiological studies on decision-making show that behaviors are chosen through coordination among multiple neural network systems, each implementing a distinct set of computational algorithms. Although these processes are commonly abnormal in neurological and psychiatric disorders, the underlying causes remain incompletely elucidated. Machine learning approaches with multidimensional data sets have the potential to not only pathologically redefine mental illnesses but also better improve therapeutic outcomes than DSM/ICD diagnoses. Furthermore, measurable endophenotypes could allow for early disease detection, prognosis, and optimal treatment regime for individuals. In this review, decision-making in real life and psychiatric disorders and the applications of machine learning in brain imaging studies on psychiatric disorders are summarized, and considerations for the future clinical translation are outlined. This review also aims to introduce clinicians, scientists, and engineers to the opportunities and challenges in bringing artificial intelligence into psychiatric practice.

Magnetic resonance imaging for individual prediction of treatment response in major depressive disorder: a systematic review and meta-analysis

No tools are currently available to predict whether a patient suffering from major depressive disorder (MDD) will respond to a certain treatment. Machine learning analysis of magnetic resonance imaging (MRI) data has shown potential in predicting response for individual patients, which may enable personalized treatment decisions and increase treatment efficacy. Here, we evaluated the accuracy of MRI-guided response prediction in MDD. We conducted a systematic review and meta-analysis of all studies using MRI to predict single-subject response to antidepressant treatment in patients with MDD. Classification performance was calculated using a bivariate model and expressed as area under the curve, sensitivity, and specificity. In addition, we analyzed differences in classification performance between different interventions and MRI modalities. Meta-analysis of 22 samples including 957 patients showed an overall area under the bivariate summary receiver operating curve of 0.84 (95% CI 0.81-0.87), sensitivity of 77% (95% CI 71-82), and specificity of 79% (95% CI 73-84). Although classification performance was higher for electroconvulsive therapy outcome prediction (n = 285, 80% sensitivity, 83% specificity) than medication outcome prediction (n = 283, 75% sensitivity, 72% specificity), there was no significant difference in classification performance between treatments or MRI modalities. Prediction of treatment response using machine learning analysis of MRI data is promising but should not yet be implemented into clinical practice. Future studies with more generalizable samples and external validation are needed to establish the potential of MRI to realize individualized patient care in MDD.

Precision psychiatry in clinical practice

The treatment of depression represents a major challenge for healthcare systems and choosing among the many available drugs without objective guidance criteria is an error-prone process. Recently, pharmacogenetic biomarkers entered in prescribing guidelines, giving clinicians the possibility to use this additional tool to guide prescription and improve therapeutic outcomes. This marked an important step towards precision psychiatry, which aim is to integrate biological and environmental information to personalise treatments. Only genetic variants in cytochrome enzymes are endorsed by prescribing guidelines, but in the future polygenic predictors of treatment outcomes may be translated into the clinic. The integration of genetics with other relevant information (e.g., concomitant diseases and treatments, drug plasma levels) could be managed in a standardised way through ad hoc software. The overcoming of the current obstacles (e.g., staff training, genotyping and informatics facilities) can lead to a broad implementation of precision psychiatry and represent a revolution for psychiatric care.Key pointsPrecision psychiatry aims to integrate biological and environmental information to personalise treatments and complement clinical judgementPharmacogenetic biomarkers in cytochrome genes were included in prescribing guidelines and represented an important step towards precision psychiatryTherapeutic drug monitoring is an important and cost-effective tool which should be integrated with genetic testing and clinical evaluation in order to optimise pharmacotherapyOther individual factors relevant to pharmacotherapy response (e.g., individual's symptom profile, concomitant diseases) can be integrated with genetic information through artificial intelligence to provide treatment recommendationsThe creation of pharmacogenetic services within healthcare systems is a challenging and multi-step process, education of health professionals, promotion by institutions and regulatory bodies, economic and ethical barriers are the main issues.

Deep learning for the prediction of treatment response in depression

BACKGROUND

Mood disorders are characterized by heterogeneity in severity, symptoms and treatment response. The possibility of selecting the correct therapy on the basis of patient-specific biomarker may be a considerable step towards personalized psychiatry. Machine learning methods are gaining increasing popularity in the medical field. Once trained, the possibility to consider single patients in the analyses instead of whole groups makes them particularly appealing to investigate treatment response. Deep learning, a branch of machine learning, lately gained attention, due to its effectiveness in dealing with large neuroimaging data and to integrate them with clinical, molecular or -omics biomarkers.

METHODS

In this mini-review, we summarize studies that use deep learning methods to predict response to treatment in depression. We performed a bibliographic search on PUBMED, Google Scholar and Web of Science using the terms "psychiatry", "mood disorder", "depression", "treatment", "deep learning", "neural networks". Only studies considering patients' datasets are considered.

RESULTS

Eight studies met the inclusion criteria. Accuracies in prediction of response to therapy were considerably high in all studies, but results may be not easy to interpret.

LIMITATIONS

The major limitation for the current studies is the small sample size, which constitutes an issue for machine learning methods.

CONCLUSIONS

Deep learning shows promising results in terms of prediction of treatment response, often outperforming regression methods and reaching accuracies of around 80%. This could be of great help towards personalized medicine. However, more efforts are needed in terms of increasing datasets size and improved interpretability of results.

Promises and pitfalls of deep neural networks in neuroimaging-based psychiatric research

By promising more accurate diagnostics and individual treatment recommendations, deep neural networks and in particular convolutional neural networks have advanced to a powerful tool in medical imaging. Here, we first give an introduction into methodological key concepts and resulting methodological promises including representation and transfer learning, as well as modelling domain-specific priors. After reviewing recent applications within neuroimaging-based psychiatric research, such as the diagnosis of psychiatric diseases, delineation of disease subtypes, normative modeling, and the development of neuroimaging biomarkers, we discuss current challenges. This includes for example the difficulty of training models on small, heterogeneous and biased data sets, the lack of validity of clinical labels, algorithmic bias, and the influence of confounding variables.

The Application of a Machine Learning-Based Brain Magnetic Resonance Imaging Approach in Major Depression

Major depressive disorder (MDD) shows a high prevalence and is associated with increased disability. While traditional studies aimed to investigate global characteristic neurobiological substrates of MDD, machine learning-based approaches focus on individual people rather than a group. Therefore, machine learning has been increasingly conducted and applied to clinical practice. Several previous neuroimaging studies used machine learning for stratifying MDD patients from healthy controls as well as in differentially diagnosing MDD apart from other psychiatric disorders. Also, machine learning has been used to predict treatment response using magnetic resonance imaging (MRI) results. Despite the recent accomplishments of machine learning-based MRI studies, small sample sizes and the heterogeneity of the depression group limit the generalizability of a machine learning-based predictive model. Future neuroimaging studies should integrate various materials such as genetic, peripheral, and clinical phenotypes for more accurate predictability of diagnosis and treatment response.

Cost-effectiveness of neuroimaging technologies in management of psychiatric and insomnia disorders: A meta-analysis and prospective cost analysis

BACKGROUND

The optimal diagnostic strategy for patients with psychiatric and insomnia disorders has not been established yet.

PURPOSE

The purpose of this study was to perform cost-effectiveness analysis of six neuroimaging technologies in diagnosis of patients with psychiatric and insomnia disorders.

METHODS

An economic evaluation study was conducted in three parts, including a systematic review for determining diagnostic accuracy, a descriptive cross-sectional study with Activity-Based Costing (ABC) technique for tracing resource consumption, and a cost-effectiveness analysis using a short-term decision-analytic model.

RESULTS

In the first phase, 93 diagnostic accuracy studies were included in the systematic review. The accumulated results (meta-analysis) showed that the highest diagnostic accuracy for psychiatric and insomnia disorders was attributed to PET (sensitivity of 90% and specificity of 80%) and MRI (sensitivity of 76% and specificity of 78%) respectively. In the second phase of the study, we calculated the cost of each technology. The results showed that MRI has the lowest cost. Based on the results in the model of cost-effectiveness sMRI ($ 50.08 per accurate diagnosis) and MRI ($ 58.54 per accurate diagnosis) were more cost-effective neuroimaging technologies.

CONCLUSION

In psychiatric disorders, no single strategy was characterized by both low cost and high accuracy. However, MRI and PET scan had lower cost and higher accuracy for psychiatric disorders, respectively. MRI was the least costly with the highest diagnostic accuracy in insomnia disorders. Based on our model, sMRI in psychiatric disorders and MRI in insomnia disorders were the most cost-effective technologies.

Machine Learning Prediction of Treatment Outcome in Late-Life Depression

Background: Recent evidence suggests that integration of multi-modal data improves performance in machine learning prediction of depression treatment outcomes. Here, we compared the predictive performance of three machine learning classifiers using differing combinations of sociodemographic characteristics, baseline clinical self-reports, cognitive tests, and structural magnetic resonance imaging (MRI) features to predict treatment outcomes in late-life depression (LLD). Methods: Data were combined from two clinical trials conducted with depressed adults aged 60 and older, including response to escitalopram (N = 32, NCT01902004) and Tai Chi (N = 35, NCT02460666). Remission was defined as a score of 6 or less on the 24-item Hamilton Rating Scale for Depression (HAMD) at the end of 24 weeks of treatment. Features subsets were constructed from baseline sociodemographic and clinical features, gray matter volumes (GMVs), or both. Three classification algorithms were compared: (1) Support Vector Machine-Radial Bias Function (SVMRBF), (2) Random Forest (RF), and (3) Logistic Regression (LR). A repeated 5-fold cross-validation approach with a wrapper-based feature selection method was used for model fitting. Model performance metrics included Area under the ROC Curve (AUC) and Matthews correlation coefficient (MCC). Cross-validated performance significance was tested by permutation analysis. Classifiers were compared by Cochran's Q and post-hoc pairwise comparisons using McNemar's Chi-Square test with Bonferroni correction. Results: For the RF and SVMRBF algorithms, the combined feature set outperformed the clinical and GMV feature sets with a final cross-validated AUC of 0.83 ± 0.11 and 0.80 ± 0.11, respectively. Both classifiers passed permutation analysis. The LR algorithm performed best using GMV features alone (AUC 0.79 ± 0.14) but failed to pass permutation analysis using any feature set. Performance of the three classifiers differed significantly for all three features sets. Important predictive features of treatment response included anterior and posterior cingulate volumes, depression characteristics, and self-reported health-related quality scores. Conclusion: This preliminary exploration into the use of ML and multi-modal data to identify predictors of general treatment response in LLD indicates that integration of clinical and structural MRI features significantly increases predictive capability. Identified features are among those previously implicated in geriatric depression, encouraging future work in this arena.

Resting State Functional Connectivity Biomarkers of Treatment Response in Mood Disorders: A Review

There are currently no validated treatment biomarkers in psychiatry. Resting State Functional Connectivity (RSFC) is a popular method for investigating the neural correlates of mood disorders, but the breadth of the field makes it difficult to assess progress toward treatment response biomarkers. In this review, we followed general PRISMA guidelines to evaluate the evidence base for mood disorder treatment biomarkers across diagnoses, brain network models, and treatment modalities. We hypothesized that no treatment biomarker would be validated across these domains or with independent datasets. Results are organized, interpreted, and discussed in the context of four popular analytic techniques: (1) reference region (seed-based) analysis, (2) independent component analysis, (3) graph theory analysis, and (4) other methods. Cortico-limbic connectivity is implicated across studies, but there is no single biomarker that spans analyses or that has been replicated in multiple independent datasets. We discuss RSFC limitations and future directions in biomarker development.

Prediction of Antidepressant Treatment Response and Remission Using an Ensemble Machine Learning Framework

In the wake of recent advances in machine learning research, the study of pharmacogenomics using predictive algorithms serves as a new paradigmatic application. In this work, our goal was to explore an ensemble machine learning approach which aims to predict probable antidepressant treatment response and remission in major depressive disorder (MDD). To discover the status of antidepressant treatments, we established an ensemble predictive model with a feature selection algorithm resulting from the analysis of genetic variants and clinical variables of 421 patients who were treated with selective serotonin reuptake inhibitors. We also compared our ensemble machine learning framework with other state-of-the-art models including multi-layer feedforward neural networks (MFNNs), logistic regression, support vector machine, C4.5 decision tree, naïve Bayes, and random forests. Our data revealed that the ensemble predictive algorithm with feature selection (using fewer biomarkers) performed comparably to other predictive algorithms (such as MFNNs and logistic regression) to derive the perplexing relationship between biomarkers and the status of antidepressant treatments. Our study demonstrates that the ensemble machine learning framework may present a useful technique to create bioinformatics tools for discriminating non-responders from responders prior to antidepressant treatments.