Artificial Intelligence and Stochastic Process-Based Analysis of Human Psychiatric Disorders

Abstract

Abstract

This paper contains an analysis and comparison of different classifiers on different datasets of Psychiatric Disorders- Personality Disorder, Depression, Anxiety, Schizophrenia and Alzheimer's disease. Psychiatric disorders are also referred to as mental disorders, abnormalities of the mind that result in persistent behavior which can seriously cause day to day function and life. Stochastic in AI refers to if there is any uncertainty or randomness involved in results and are used during optimization; Using this process also helps to provide precise results. The study of stochastic process in AI uses mathematical knowledge and techniques from probability, set theory, calculus, linear algebra and mathematical analysis like Fourier analysis, real analysis, and functional analysis. this technique is used to construct neural network for making artificial intelligent mode for processing and minimizing human effort. This paper contains classifiers like SVM, MLP, LR, KNN, DT, and RF. Several types of attributes are used and have been trained by Weka tool, MATLAB, and Python. The results show that the SVM classifier showed the best performance for all the attributes and disorders researched in this paper.

Full text

JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Yashvi Bhavsar1, Khyati Mistry1, Nishchay Parikh1, Himani Shah1, Adarsh Saraswat1, Helia Givian2, Mojataba Barzegar3,4,5, Maryam Hosseini6, Khojaste
Rahimi Jaberi6, Archana Magare1, Mohammad Javad Gholamzadeh7, Hadi Aligholi6, Ali-Mohammad Kamali6, Prasun Chakrabarti8, Mohammad Nami5,6,9*
1. Department of Computer Science and Engineering, Institute of Technology & Management Universe, Dhanora Tank Road, Near Jarod, Vadodara - 391510,
Gujarat, India
2. Preclinical Core Facility, Tehran University of Medical Sciences, Tehran, Iran
3. Intelligent quantitative biomedical imaging (iqbmi), Tehran, 1955748171, Iran
4. School of Medical Physics and Medical Engineering, Tehran University of Medical Sciences, Tehran, 14399-55991, Iran
5. Society for Brain Mapping and erapeutics (SBMT), Brain Mapping Foundation (BMF), Middle East Brain + Initiative, Los Angeles, CA 90272, CA, USA
6. Department of Neuroscience, School of Advanced Medical Sciences and Technologies, Shiraz University of Medical Sciences, Shiraz, 71348-14336, Iran
7. Students' Research Committee, School of Medicine, Shiraz University of Medical Sciences, Shiraz, Iran
8. Provost, Techno India JNR, Institute of Technology, Udaipur 313003, Rajasthan, India
9. Neuroscience Center, Instituto de Investigaciones Cientícas y Servicios de Alta Tecnología (INDICASAT AIP), City of Knowledge, Panama City, Republic of
Panama
Use your device to scan and
read the article onlinne Citaon
Bhavsar Y, Mistry K, Parikh N, Shah H, Saraswat A, Givian H, Barzegar M, Hosseini M, Rahimi Jaberi K, Magare A,
Gholamzadeh MJ, Aligholi H, Kamali AM, Chakrabarti P, Nami M. Articial intelligence and stochastic process-based analysis of human
psychiatric disorders. JAMSAT. 2021; 6(1): 33-53.
https://dx.doi.org/10.30476/jamsat.2021.93288.1027
A B S T R A C T
is paper contains an analysis and comparison of dierent classiers on dierent datasets
of Psychiatric Disorders- Personality Disorder, Depression, Anxiety, Schizophrenia
and Alzheimer's disease. Psychiatric disorders are also referred to as mental disorders,
abnormalities of the mind that result in persistent behavior which can seriously cause day
to day function and life. Stochastic in AI refers to if there is any uncertainty or randomness
involved in results and are used during optimization; Using this process also helps to
provide precise results. e study of stochastic process in AI uses mathematical knowledge
and techniques from probability, set theory, calculus, linear algebra and mathematical
analysis like Fourier analysis, real analysis, and functional analysis. this technique is used
to construct neural network for making articial intelligent mode for processing and
minimizing human eort. is paper contains classiers like SVM, MLP, LR, KNN, DT,
and RF. Several types of attributes are used and have been trained by Weka tool, MATLAB,
and Python. e results show that the SVM classier showed the best performance for all
the attributes and disorders researched in this paper.
Keywords:
Alzheimer’s disease, Anxiety,
Articial Intelligence,
Depression, DT, KNN,
Logistic Regression, MLP,
Personality Disorder, RF,
Schizophrenia, SVM
Article info:
Received: 17 October 2021
Accepted: 06 December 2021
* Corresponding Author:
Mohammad Nami, MD, PhD.
Address: Department of Neuroscience, School of Advanced Medical Sciences and Technologies, Shiraz University of Medical Sciences, Shiraz, Iran
Tel: +987132305488
E-mail: torabinami@sums.ac.ir
33
December 2021, Volume 6, Issue 1
Research Paper: Articial intelligence and stochastic process-
based analysis of human psychiatric disorders

34
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
1. Background
Psychiatric Disorder is a mental illness that
disturbs one’s thinking, moods, and behaviors. Its
consequence is an increase in risk of disability, pain,
death, or loss of freedom. Recently, the implementation
areas of Articial Intelligence (AI) are growing. AI
and computer simulation play a vital role in domain
research such as statistics, forecasting, discovery,
and more. With this technology, a solution to various
problems can be found accordingly. Moreover, the
stochastic process is when the initial state is known;
however, the next state is unpredictable. Also, it has
randomness, and the patterns need to be recognized.
In this paper, diverse psychiatric disorders’ datasets
are analyzed for their ecacy in predicting psychiatric
disorders. Furthermore, the Weka tool, MATLAB,
and Python are used to perform the analyses. The
selection of which classiers to use was based on
the research done in the literature survey. Depression
is a constant feeling of sadness and loss of interest.
A personality disorder is a mental disorder in which
one has a rigid and unhealthy pattern of thinking,
functioning, and behaving. Anxiety disorder is
excessive fear or stress. Schizophrenia is a serious
mental illness that causes irrational thoughts,
abnormal actions, delusion and wandering, such as
hearing loss.
Alzheimer’s disease (AD) is a progressive and
degenerative brain disorder that results in nerve cell
death, tissue loss, and memory loss in the brain. The
global rate of AD is gradually increasing, and early
diagnosis of AD is essential for the patient’s care to
control and prevent the progression of the disease for
future treatments.
2. Literature Survey
Table 1 summarizes methodology and ndings of
studies regarding psychiatric disorders.
3. Method
Classication is a technique to identify the classes
of given data points. To classify the data, we use
classiers. Classiers take some data as input
(training data), using which they train the model
and nd out how the data are related according to
a class. Five disorders (Depression, Personality
Disorder, Anxiety, Schizophrenia, and Alzheimer’s
Disease) are explained in this paper. One dataset for
each disorder is taken for training and analysis. We
focused on features from every dataset and classiers
were applied on each of the attributes. Analysis of
each dataset by selecting the most important attribute
out of many and analyzing them with classiers, is
the main process of our dataset analysis. Based
on research we have listed, classiers who have
achieved highest accuracy in psychiatric disorders,
on reference to this, Support Vector Machine (SVM),
Logistic Regression, Multi-Layer Perceptron (MLP),
Decision Tree (DT), k-Nearest Neighbor (KNN) and
Random Forest (RF), are chosen as shown in Figure 1.
Support Vector Machine (SVM)
A Support Vector Machine is a machine learning
algorithm used mainly for classication. SVM works
both for regression and classication, but nowadays
it is mostly used for classication due to its precise
classifying ability. In this, the data is divided into
n-dimensional spaces called hyperplanes. A decision
boundary is created so that the new data can be
separated and put according to its features into
dierent classes (10).
Logistic Regression (LR)
Logistic Regression is the extension of linear
regression. In this, instead of straight lines or
hyperplanes, data is tted using a logistic function
with only two possible outcomes, 0 and 1 (12).
Multi-Layer Perceptron (MLP)
MLP is an addition to the feed-forward neural
network. In this, there are three layers- Input layer
(which receives the input data to be preprocessed),
output layer (which performs the classication) and
hidden layer present between input and output layer,
containing arbitrary layers that are true computational
engines of MLP (12).
Decision Tree (DT)
A Decision Tree is a owchart type tree-like structure
whose internal nodes are the tests on attributes, each
leaf node holds a class label and every branch is the
outcome of the test (12).
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

35
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Arribas et al.
“A signature-based machine learning
model for distinguishing bipolar
disorder and borderline personality
disorder” (2018). (1)
- 130 participants (BD, BDP, and
healthy individuals)
- Classifier: Random forest
- Signature-Based Model
Mood scores accuracy:
- Healthy = 89–98%
- BD = 82–90%
- BDP = 70–78%
Acharya et al.
“Automated EEG-based Screening of
Depression Using Deep Convolutional
Neural Network” (2018). (2)
- 30 Participants (15 depressed and
fifteen healthy individuals)
- EEG signals of both left and right
brain hemispheres (open and rest
states 5 minutes)
- 13 layered Deep CNN model
- Ninety percent train, 10% test
- The network was trained using the
backpropagation algorithm with a
batch size of five.
Left hemisphere:
- Accuracy = 93.54%,
- Sensitivity = 91.89%,
- Specificity = 95.18%.
Right hemisphere:
- Accuracy = 95.49%,
- Sensitivity = 94.99%,
- Specificity = 96.00%.
Shrivastava et al.
“A SVM-based classification approach
for obsessive compulsive disorder by
oxidative stress biomarkers” (2019).
(3)
- OCDP and Non-OCDP participants
- Blood samples
- SVM with 2 variants, Random
Forest, Linear Discriminant Analysis,
and K-NN (base classifier)
- 5 osbMarkers
- Clusters: K-Means and Fuzzy C-
Means
- SVMR accuracy = 97±1%.
- Fuzzy C-Means better with accuracy
= 76.67%
Saeedi et al.
“Major depressive disorder
assessment via enhanced k‑nearest
neighbor method and EEG signals”
(2020). (4)
- 34 depressed and thirty healthy
individuals
- Feature selection method:
Genetic Algorithm
- Three Classifiers: E-KNN, SVM, and
MLP
- 10-Cross Validation
Accuracy were as follows:
- E-KNN = 98.44% (±3.4), SVM =
92.18% (±6.9), KNN = 95.31% (±5.2),
MLP = 93.75% (±6.8)
- Sensitivity: E-KNN = 97.10%, SVM =
88.23%, KNN =96.80%, MLP =
90.00%
- Specificity: E-KNN = 100%, SVM
=96.66%, KNN =93.33% MLP
=94.41%
Cremers et al.
“Borderline personality disorder
classification based on brain network
measures during emotion regulation”
(2020). (5)
- Participants: 51 BPD, 26 Cluster C
Disorder, and forty-four non-patients
- fMRI data (acquired and
preprocessed)
- Images formatted from BrainVoyager
to nifti format.
- Phasic and Tonic Networks
- Classifier: Linear Support Vector
Machine
- Tonic-Strength model highest
Balanced Accuracy = 62%
- BPD vs NPC = 55% (bal. acc.)
Table 1. Literature survey on psychiatric disorders
December 2021, Volume 6, Issue 1

36
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Yang et al.
“Multivariate classification of drug-
naive obsessive-compulsive disorder
patients and healthy controls by
applying an SVM to resting-state
functional MRI data” (2019). (6)
- Examined fractional Amplitude Low-
Frequency Fluctuation (fALFF)
- Applied Support Vector Machine
(SVM) to discriminate OCD patients
- Values of fALFF, calculated from
sixty-eight drug-naive OCD patients
and sixty-eight demographically
matched healthy controls
(classification model)
- SVM = 72%
Khazbak et al.
“MindTime: Deep Learning Approach
for Borderline Personality Disorder
Detection” (2021). (7)
- User adds a diary input.
- Analyzed to detect if there are signs
of BPD symptoms.
- Investigation of different classifiers
to extracts features (Naive Bayes,
SVM, KNN, and finally LSTM)
- SVM = 90.1%
- LSTM = 91%
- CNN = 65%
Bracher-Smith et al.
“Machine learning for genetic
prediction of psychiatric disorders: a
systematic review” (2021). (8)
- Classifiers: naive Bayes, k-Nearest
Neighbors (k-NN), penalized
regression, decision trees, random
forests, boosting, Bayesian networks,
Gaussian processes, Support Vector
Machines (SVMs), and neural
networks
- Dataset: thirteen studies were
selected for inclusion, containing
seventy-seven distinct ML models
- Type: psychiatric disorders from
genetics alone
- Schizophrenia :0.54–0.95 AUC
- Bipolar: 0.48–0.65 AUC
- Autism: 0.52–0.81 AUC
- Anorexia: 0.62–0.69 AUC
- AUC: Area Under the ROC (Receiver
Operating Characteristic) Curve
Saidi et al.
“Hybrid CNN-SVM classifier for
efficient depression detection
system” (2021). (9)
- 189 participants (audio): 107
training set, thirty-five validation set,
and forty-seven test set
- DAIC-WOZ dataset is used.
- Using CNN classifier for training
- The fully connected layers are
replaced by SVM layers.
- Input map is given to CNN.
- The classification is done by an SVM
classifier.
Accuracy:
- CNN = 58.57%
- Hybrid CNN-SVM (the proposed
model) = 68%
Priya et al.
“Predicting Anxiety, Depression and
Stress in Modern Life using Machine
Learning Algorithms” (2019). (10)
- 348 participants
- DT, Random Forest, Naive Bayes,
SVM, and KNN.
- Data from the Depression, Anxiety
and Stress Scale questionnaire (DASS
‘21)
- Scores labeled on the basis of
severity - normal, mild, moderate,
severe, and extremely severe.
For depression:
- Accuracy
- DT=0.778, RF=0.798, Naive Bayes=
0.855, SVM=0.803, KNN=0.721
- F1 Score
- DT=0.723, RF=0.766, Naive Bayes=
0.836, SVM=0.765, KNN=0.687
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

37
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Pan et al.
“Detecting Manic State of Bipolar
Disorder Based on Support Vector
Machine and Gaussian Mixture Model
Using Spontaneous Speech” (2018).
(11)
- Two classifiers are used- SVM
(Support Vector Machine) and GMM
(Gaussian Mixture Model)
- Data Collection: twenty-one
hospitalized patients’ speeches were
recorded.
- Features Extraction: key features
pitch, MFCC, etc. extracted by
software SMILE.
The LIBSVM toolbox → SVM
HTK tool → GMM
- 3 patients for single patient test &
21 patients for multiple patients test
- The manic state detection
accuracies of SVM and GMM
compared using student's t-test,
For single patient experiment,
accuracy obtained overall:
- SVM = 88.56±5.26
- GMM = 84.46±1.85
- For multiple patient experiments,
accuracy overall:
- SVM =60.87±18.90
- GMM = 72.27±6.90
Dabhane et al.
“Depression Detection on Social
Media using Machine Learning
Techniques” (2021). (12)
- Logistic Regression, KNN, SVM, DT,
MLP, and Naive Bayes
- Data collection: diverse types of
tweets from twitter API (CSV file)
- Data Preprocessing: removal of
duplicate entries
- Exploratory Data Analysis: analyze
datasets and collect key features
- Training and Testing 2 steps:
1. Implementing Algorithms
Individually
2. Implementing Ensemble Learners:
Here, the voting classifier and
Blending ensemble classifier were
used for greater performance and
accuracy.
- KNN = 73.29%
- Logistic regression = 84.86%
- SVM = 85.04%
- Naive Bayes Classifier = 83.04%
- Decision tree=80.53%
- Multilayer perceptron (MLP) =
78.65%
For ensemble implementation:
- Voting Classifier = 85.35%
- Blending Classifier = 87.21%
Islam et al.
“Detecting Depression Using K-
Nearest Neighbors (KNN)
Classification Technique” (2018). (13)
- 7145 Facebook comments data was
collected using NCapture and
processed using the LIWC2015 tool
and then paraphrases were extracted
to detect emotions.
- Different KNN classifiers like Fine
KNN, Medium KNN, Coarse KNN,
Cosine KNN, Cubic KNN, and
Weighted KNN were applied and their
f-measure was compared.
- The experiment was done in 10-fold
cross-validation on all testing
datasets.
Emotional Process:
- Fine KNN = 0.59, Medium KNN =
0.59, Coarse KNN = 0.71, Cosine
KNN = 0.58, Cubic KNN = 0.59,
Weighted KNN= 0.60
Linguistic Style:
- Fine KNN = 0.58, Medium KNN =
0.57, Coarse KNN = 0.70, Cosine
KNN = 0.60, Cubic KNN = 0.57,
Weighted KNN= 0.62
Temporal Process:
- Fine KNN = 0.58, Medium KNN =
0.57, Coarse KNN = 0.70, Cosine
KNN = 0.59, Cubic KNN = 0.57,
Weighted KNN= 0.58
All features:
- Fine KNN = 0.58, Medium KNN =
0.56, Coarse KNN = 0.67, Cosine
KNN = 0.60, Cubic KNN = 0.55,
Weighted KNN= 0.61
December 2021, Volume 6, Issue 1

38
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Wang et al.
“Using Electronic Health Records and
Machine Learning to Predict
Postpartum Depression (PPD)”
(2019). (14)
- Clinical assessment of PPD was
used as the outcome based on
Statistics Canada and International
Classification of Diseases (ICD-10-
CM) .
- AI methods included LR, SVM, DT,
NB, XGBoost, and RF.
The results suggest a potential for
applying machine learning to EHR
data to predict PPD and inform
healthcare delivery.
Best prediction performance achieved
an AUC of 0.79 that it was for SVM
model.
Al-ezzi et al.
“Severity Assessment of Social
Anxiety Disorder Using Deep Learning
Models on Brain Effective
Connectivity” (2021). (15)
- They recruited eighty-nine
participants from 502.
- Examine the SAD data and develop a
task for SAD assessment to acquire
EEG data.
- EEG data preprocessing is done. And
high and low frequency deflections.
- Dataset helps to explain brain
activity values.
- Applied connectivity features for
precision based SAD prediction based
on PDC algorithm.
- Different Deep learning network
algorithm applied and then do
analysis on the findings.
The CNN+LSTM is more accurate than
the 2 layer or 3 layer CNN and LSTM
with attention mechanisms.
The result of CNN+LSTM:
- Accuracy=93%
- Sensitivity=95%
- Specificity=85%
- Precision=86%
Sau and Bhakta,
”Screening of anxiety and depression
among seafarers using machine
learning technology” (2019). (16)
- Study the variable data including the
questionnaire with people such as
their age, education, family, etc.
- Feature selection eliminates the
irrelevant features from the set of
predictor values.
- A final dataset with all fourteen
features and one target and 470
instances were prepared for different
classification. This was divided into
two groups based on the period of
data collection.
The classifiers and their accuracy and
precision data:
- CatBoost = 89.3%, 89%
- Logistic regression = 87.5%, 84%
- SVM = 82.1%, 80.7%
- Naive Bayes = 82.1%, 76.9%
- Random Forest = 78.6%, 80.7%
Jothi et al,
”Predicting generalized anxiety
disorder among women using Shapley
value” (2020). (17)
- The data acquisition phase, data
relevant to the study were collected.
- After that, Data cleaning and
transformation would be performed.
- The feature selection using Shapley
value was conducted using the
original GAD features.
- selected features were used as
inputs for classification prediction
algorithms.
- In the last phase, classification
performance criteria were used to
evaluate the prediction algorithm.
The performance of prediction models
without feature selection is less than
with the feature selection.
The accuracy, sensitivity and
specificity of classifier with feature
selection, respectively:
- Naive Bayes = 80%, 98.79%,
76.47%
- Random Forest = 90.6%, 93.47%,
69.53%
- J48 = 95.70%, 97.5%, 86.3%
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

39
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Gui et al.
“The Impact of Emotional Music on
Active ROI in Patients with Depression
Based on Deep Learning: A Task-
State fMRI Study” (2019). (22)
- A large convolution kernel of the
same size as the correlation matrix
for the feature matching of 264 ROIs.
1. 4D fMRI data are used to generate
the 2D correlation matrix of one
person’s brain based on ROIs
2. processed by the threshold value
which is selected according to the
characteristics of complex network
and small-world network. After that,
the DLM in this paper is compared
with SVM, logistic regression (LR), k-
Nearest Neighbor (kNN), a common
DNN, and a deep CNN for
classification.
3. Calculate the matched ROIs from
the intermediate results of the DLM
which can help related fields further
explore the pathogeny of depression
patients.
Deep analysis of the brain
mechanism of depressed patients is
more conducive to solving the
condition of depressed patients.
Wen et al.
”Deep Learning Methods to Process
fMRI Data and Their Application in the
Diagnosis of Cognitive Impairment”
(2018). (23)
- DL methods in fMRI Data Analysis:
CNN (Feature Extraction, Auto-
Encoder, 3D-CNN); FNN;
- Development of DL Methods for
fMRI Data Analysis in Cognitive
Impairment.
This study reviewed the recent
literature of deep learning used in
fMRI data.
We can make full use of the auto-
extracted features to improve
accuracy of deep learning methods.
Liu et al.
“Classification of Alzheimer’s Disease
by Combination of Convolutional and
Recurrent Neural Networks Using
FDG-PET Images” (2018). (24)
- 3D FDG-PET image;
- ADNI dataset( PET-MRI-Othertests);
- MCI, NC and early AD classification.
- Using deep 2D CNN network,
Recurrent neural networks (RNNs);
- BGRU network layer for classification
BGRU can boost the classification.
This method performs better than
others.
Sau and Bhakta,
“Predicting anxiety and depression in
elderly patients using machine
learning technology” (2017). (25)
- 510 participants
- Ten classifiers (BN, NB, Log, MLP,
SMO, KS, RS, J48, RF, RT) were
evaluated with a data set of geriatric
patients.
- They were tested with a 10-fold
cross validation method.
The results showed that Random
forest predicts anxiety and depression
in elderly patients better than other
classifiers also with accuracy 91%
and false positive 10%, gold standard
tool.
- RF (AUC: 94.3, Accuracy: eighty-nine,
F1: 85.1)
McGinnis et al.
“Rapid Anxiety and Depression
Diagnosis in Young Children Enabled
by Wearable Sensors and Machine
Learning” (2018). (26)
- 63 children and their primary
caregivers
- DSM-IV was checked to diagnose
mental disorders.
- Participant motion was tracked using
a belt-worn IMU.
- classification accuracy compared for
SVM, DT, kNN, LR also for just
accelerometer features (ACC), just
gyro features (GYR), just angle
features (ANG).
Analysis suggests that, when paired
with machine learning, 20 seconds of
wearable sensor data extracted from
a fear induction task can be used to
diagnosis internalizing disorders in
young children with a high level of
accuracy and at a fraction of the cost
and time of existing assessment
techniques and the LR model is the
best performing compared to other
with accuracy of 80% and AUC 0.92.
December 2021, Volume 6, Issue 1

40
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Gui et al.
“The Impact of Emotional Music on
Active ROI in Patients with Depression
Based on Deep Learning: A Task-
State fMRI Study” (2019). (22)
- A large convolution kernel of the
same size as the correlation matrix
for the feature matching of 264 ROIs.
1. 4D fMRI data are used to generate
the 2D correlation matrix of one
person’s brain based on ROIs
2. processed by the threshold value
which is selected according to the
characteristics of complex network
and small-world network. After that,
the DLM in this paper is compared
with SVM, logistic regression (LR), k-
Nearest Neighbor (kNN), a common
DNN, and a deep CNN for
classification.
3. Calculate the matched ROIs from
the intermediate results of the DLM
which can help related fields further
explore the pathogeny of depression
patients.
Deep analysis of the brain
mechanism of depressed patients is
more conducive to solving the
condition of depressed patients.
Wen et al.
”Deep Learning Methods to Process
fMRI Data and Their Application in the
Diagnosis of Cognitive Impairment”
(2018). (23)
- DL methods in fMRI Data Analysis:
CNN (Feature Extraction, Auto-
Encoder, 3D-CNN); FNN;
- Development of DL Methods for
fMRI Data Analysis in Cognitive
Impairment.
This study reviewed the recent
literature of deep learning used in
fMRI data.
We can make full use of the auto-
extracted features to improve
accuracy of deep learning methods.
Liu et al.
“Classification of Alzheimer’s Disease
by Combination of Convolutional and
Recurrent Neural Networks Using
FDG-PET Images” (2018). (24)
- 3D FDG-PET image;
- ADNI dataset( PET-MRI-Othertests);
- MCI, NC and early AD classification.
- Using deep 2D CNN network,
Recurrent neural networks (RNNs);
- BGRU network layer for classification
BGRU can boost the classification.
This method performs better than
others.
Sau and Bhakta,
“Predicting anxiety and depression in
elderly patients using machine
learning technology” (2017). (25)
- 510 participants
- Ten classifiers (BN, NB, Log, MLP,
SMO, KS, RS, J48, RF, RT) were
evaluated with a data set of geriatric
patients.
- They were tested with a 10-fold
cross validation method.
The results showed that Random
forest predicts anxiety and depression
in elderly patients better than other
classifiers also with accuracy 91%
and false positive 10%, gold standard
tool.
- RF (AUC: 94.3, Accuracy: eighty-nine,
F1: 85.1)
McGinnis et al.
“Rapid Anxiety and Depression
Diagnosis in Young Children Enabled
by Wearable Sensors and Machine
Learning” (2018). (26)
- 63 children and their primary
caregivers
- DSM-IV was checked to diagnose
mental disorders.
- Participant motion was tracked using
a belt-worn IMU.
- classification accuracy compared for
SVM, DT, kNN, LR also for just
accelerometer features (ACC), just
gyro features (GYR), just angle
features (ANG).
Analysis suggests that, when paired
with machine learning, 20 seconds of
wearable sensor data extracted from
a fear induction task can be used to
diagnosis internalizing disorders in
young children with a high level of
accuracy and at a fraction of the cost
and time of existing assessment
techniques and the LR model is the
best performing compared to other
with accuracy of 80% and AUC 0.92.
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

41
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
McGinnis et al.
“Giving Voice to Vulnerable Children:
Machine Learning Analysis of Speech
Detects Anxiety and Depression in
Early Childhood” (2019). (27)
- 71 children who spoke fluent English
and their caregivers
- Using DSM-IV to diagnosis children
with internalizing.
- Assessment of audio features to
characterize the ability of the
proposed approach for identifying
children with an internalizing disorder
- Classification models included LR,
SVM with a linear and Gaussian
kernel and RF.
The results showed that machine
learning analysis of audio data from
the task can be used to identify
children with an internalizing disorder
with 80% accuracy (54% sensitivity,
93% specificity).
This new tool is shown to outperform
clinical thresholds on parent-reported
child symptoms, which identify
children with an internalizing disorder
with lower accuracy and similar
specificity and sensitivity in this
sample.
Nemesure et al.
“Predictive modeling of depression
and anxiety using electronic health
records and a novel machine learning
approach with artificial intelligence”
(2021). (28)
- Use of Electronic Health Records
(EHR) data of 4184 undergraduate
students
- A total of fifty-nine biomedical and
demographic features from the
general health survey were used.
- Psychiatric diagnoses were done by
a multi-stage process such as using
DSM-IV
- AI methods included XGBoost, RF,
SVM, kNN and NN
The results indicated moderate
predictive performance for the
application of machine learning
methods in detection of GAD and
MDD based on EHR data.
Richter et al.
“Using machine learning-based
analysis for behavioral differentiation
between anxiety and depression”
(2020). (29)
- 125 participants: included HA, HD,
HAD, and LAD (control)
- Questionnaires to assess anxiety
and depression included DASS-213,
STAI-T27, BDI–II28, RRS29, and
PSWQ30.
- Behavioral tasks included EDPT,
RTs, FAFT, WIT, WSAP, FET, and IST
- AI method: DT
The prediction model for
differentiating between symptomatic
participants (i.e., high symptoms of
depression, anxiety, or both)
compared to control revealed a
71.44% prediction accuracy for the
former (sensitivity) and 70.78% for
the latter (specificity). 68.07% and
74.18% prediction accuracy was
obtained for a two-group model with
high depression/anxiety, respectively
and Distinguishing between anxiety
and depression by specific behavioral
measures.
Chen et al.
“Detecting Abnormal Brain Regions in
Schizophrenia Using Structural MRI
via Machine Learning” (2020). (30)
- Sample size: COBRE: Paranoid
SZ=34, NC=34
- Extraction white matter and gray
matter volume
- Using SVM classifier
- Accuracy = 85.27%
- Sensitivity = 85.87%
- Specificity = 85.08%
Calhas et al.
”On the use of pairwise distance
learning for brain signal classification
with limited observations” (2020).
(31)
- A sample of eighty-four people;
Feature extraction was performed
using SNN architecture along with
DSTFT.
- After receiving the output of feature
extraction the following classifiers
were trained SVM, RF, XGB, NB, and
KNN. This process was performed in
LOOCV.
From the tested classifiers. DSTFT-
SNN-XGB were found to be the most
efficient.
- Accuracy = 0.95±0.05%
- Sensitivity = 0.98±0.02%
- Specificity = 0.92±0.07%
December 2021, Volume 6, Issue 1

42
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Fernando et al.
“Neural memory plasticity for medical
anomaly detection” (2020). (32)
- Recurrent ANN: LSTM layers
followed by a Neural Memory Network
with plasticity mechanism using EEG
recordings from the auditory oddball
trials.
There are no existing machine
learning models that attempt the
classification of schizophrenia risk
using EEGs. With 93.86 ± 0.21%
accuracy.
Guo et al.
”Support Vector Machine-Based
Schizophrenia Classification Using
Morphological Information from
Amygdaloid and Hippocampal
Subregions” (2020). (33)
- Sample size: COBRE: SZ=179,
NC=77
- Extraction structural features
(hippocampus, amygdala)
- Using SVM classifier
- Accuracy = 81.75%
- Sensitivity = 84.21%
- Specificity = 81.16%
Phang et al.
”A Multi-Domain Connectome
Convolutional Neural Network for
Identifying Schizophrenia From EEGs
Connectivity Patterns” (2020). (34)
- The proposed approach uses the
MDC-CNN framework for classifying
SZ and Healthy Control (HC) using
EEG based effective brain networks.
- Feature extraction was based on
Time domain VAR model coefficient
matrix (2D), frequency domain PDC
matrix (2D), and hand crafted
complex network measures (1D).
Performance of MDC-CNN on decision
level
- Accuracy = 91.69%
- Sensitivity = 91.11%
- Specificity = 92.50%
- Classification time= 0.81s
Oh et al.
“Identifying Schizophrenia Using
Structural MRI With a Deep Learning
Algorithm”, (2020). (35)
-- Sample size: BrainGluSchi COBRE,
MCICShare, MorphCH, NUSDAST:
SZ=443, NC=423
- Normalization
- Create 3D images
- Divide the brain into eight regions in
each image
- Using 3D CNN for classifier
Acc=97
Sen=96
Spec=96
Matsubara et al.
“Deep Neural Generative Model of
Functional MRI images for Psychiatric
Disorder Diagnosis” (2019). (36)
- The proposed technique accepts any
type of fMRI time series. It can be a
3D, 2D, k-space image, a vector of
voxels, a feature vector of ROIs or a
state of dynamic functional
connectivity.
- The DGM (deep neural generative
model) approach was implemented
using deep neural networks.
The accuracy of the proposed model
for the following disorder are:
SZ=71.3%
BD=64%
Talpalaru et al.
”Identifying schizophrenia subgroups
using clustering and supervised
learning” (2019). (37)
- Sample size: NUSDAST: SZ=104,
NC=63
- Segmentation using CIVET pipeline
- Extraction means cortical thickness
value from seventy-eight regions
- Using agglomerative hierarchical
clustering to feature
reduction/selection
- Using SVM, RF, Logistic regression to
prediction that RF was better than
others (accuracy)
Acc=75
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

43
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
RReesseeaarrcchh TTiittllee
MMeetthhooddoollooggyy
FFiinnddiinnggss
Liang et al,
“Classification of First-Episode
Schizophrenia Using Multimodal Brain
Features: A Combined Structural and
Diffusion Imaging Study” (2019). (38)
- This paper discussed identifying
schizophrenia and multimodal
multivariate neuroimaging features.
- Multiple brain measures including
regional Gray Matter (GM) volume,
cortical thickness, gyrification,
Fractional Anisotropy (FA), and Mean
Diffusivity (MD) were extracted using
fully automated procedures.
- Gradient Boosting Decision Tree was
then applied on the structural MRI
data.
- 75.05% Accuracy was achieved from
fused structural and diffusion tensor
imaging metrics.
- Average accuracy derived from
combined features selected from
cortical thickness, gyrification, FA, and
MD was 76.54%.
- 63.50% for GMV, 66.47% for cortical
thickness, and 66.00% for MD. In
another dataset, average accuracy
was 54.70% for GMV, 60.94% for
cortical thickness, and 67.43% for
MD.
Chatterjee et al,
”Identification of brain regions
associated with working memory
deficit in schizophrenia” (2019). (39)
- Preprocessing of functional MRI
data.
- Group ICA is applied to the Time
series fMRI data.
- Segment ICs with AAL atlas. Then
extracting statistical features for each
segment.
- Applying FDR for feature ranking and
classification using the feature
subsets for each IC in LOOCV (leave-
one-out cross validation) and SVM,
and k-nearest neighbors.
Ninety-four percent (SVM)
Ninety-six percent (1-NN)
Kalmady et al.
”Towards artificial intelligence in
mental health by improving
schizophrenia prediction with multiple
brain parcellation ensemble-learning”
(2019). (40)
-Firstly, image acquisition of MRI data
was done. Then image pre-processing
was performed, in which pre-
processing and feature extraction was
done using MATLAB. After that each
functional image was smoothed using
a 4mm FWHM Gaussian kernel.
Lastly, prediction and evaluation
framework. “L2-regularized Logistic
regression” AI technique was used.
The Accuracy of the L2 regularized
logistic regression technique is 87%.
Qureshi et al.
”3D-CNN based discrimination of
Schizophrenia using resting-state
fMRI” (2019). (41)
Structural data acquisition, functional
data acquisition, pre-processing of
functional MRI data, independent
component analysis using MELODIC,
classification using 3D-CNN deep
learning framework.
Acc=98.01%
Sen=97.49%
Spec=98.62%
Yu et al.
“Magnetic resonance imaging study
of gray matter in schizophrenia based
on XGBoost” (2018). (42)
- Sample size: Clinical: SZ=100,
NC=100
- Extraction GLCM features
- Using XGBoost classifier
Acc=72
Manohar and Ganesan.
”Diagnosis of Schizophrenia Disorder
in MR Brain Images Using Multi-
objective BPSO Based Feature
Selection with Fuzzy SVM” (2018).
(43)
- Sample size: NAMIC: 60 Images
from 20 people (SZ+NC)
- Extraction hu moments, GLCM,
zernike moments, and structure
tensor
- Using BPSO based on fuzzy SVM
classifier
Acc=90
Sen=92.86
Spec=87.5
December 2021, Volume 6, Issue 1

44
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
Random Forest (RF)
Random forest (RF) models are machine learning
models that make output predictions by combining
outcomes from a sequence of regression decision
trees (60). It creates many classication trees and
a bootstrap sample technique is used to train each
tree from the set of training data. This method only
searches for a random subset of variables in order
to obtain a split at each node. For the classication,
the input vector is fed to each tree in the RF, and
each tree votes for a class. Finally, the RF chooses
the class with the highest number of votes (61).
k-Nearest Neighbor (KNN)
KNN assumes that similar things exist nearby.
KNN classies the new data into most related
categories. It collects and stores all the available
data and then classies a new category based on
similarities between data (12).
3.1. Proposed Method
3.1.1. Weka Tool Approach
Weka tool is a collection of data mining tasks with
machine learning algorithms. Data prepossessing,
regression, visualization, classication, clustering,
and association rules are predened tools in the
Weka tool. In this paper, for depression, personality
disorder, anxiety, and schizophrenia, we have used
the weka tool to apply three classiers; SVM,
Logistic, and MLP. In this approach, dataset analysis
is done by Preprocessing and Classifying the dataset
attributes. A owchart of this approach is shown in
Figure 2.
Raw dataset is taken as input in the weka tool.
Instances store all values (nominal, numeric) in
oating point numbers, if the attribute is nominal
then the value is stored at the corresponding value
in attributes denition. Every dataset had dierent
instances. In Weka tool the rst step is to preprocess
the dataset and then classify it to obtain results.
1. Preprocessing
Attributes are the elds of data. They are also called
features of the data. Attributes in the dataset can have
data types such as: numeric (contains a oating-point
number), nominal (represents a xed set of nominal
values), string (represents a dynamically expanding
set of nominal values), date (represents a date),
and relational (contain other attributes) is used for
representing Multi-Instance data. In preprocessing,
rstly nominal attributes are normalized after
discretization is applied to the same. Secondly,
numeric attributes are standardized for setting up the
standard deviation to one.
Figure 1. Comparison of classiers used in literature
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

45
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
a. Normalization
This method allows the transformation of any
element from an equivalence class of any shape
transforms into a specic one. It helps to eliminate the
gross inuences. Dataset was raw before normalizing,
after normalizing all the nominal attributes of the
dataset, the minimum and maximum values were set
to 0 and 1 respectively. The scale is set to 1.0. The
value of distinct was forty-four. This is how attribute
values are brought to alignment using normalization.
b. Discretization
This lter allows converting a real-valued attribute
into an ordinal attribute. It is a process of dividing
the geometry of a dataset into nite elements to
prepare for analysis. Dataset was normalized before
applying discretization to attributes. After applying
discretization to all nominal attributes, the value of
count and weight became the same. The labels were
divided into ten parts having a bin range precision of
six. The desired weight of instances per interval was
1.0. Value of distinct became ten from forty-four after
discretization.
c. Standardization
This lter is a scaling technique where the values
are centered around the mean with a unit standard
deviation. Unique standard deviation is obtained by
standardization. Before applying standardization the
standard deviation, mean, maximum and minimum
can be any oat value. After applying this method,
the value of standard deviation becomes 1.0 for
every numeric attribute in the dataset. Minimum and
maximum values can be any negative/positive oat
value while the mean will be set to 0. Preprocessing
is complete and the dataset is ready to classify.
2. Classication
This method includes classier, attribute
selection, and test options. Test options are used
to set the percentage split value or cross-validation
(folds). Only one attribute can be chosen at a time
to obtain specic results for that attribute. Folds in
machine learning means the distribution of data into
equivalent parts like threefold, four-fold, etc.
Figure 2. Weka Tool Dataset Analysis
December 2021, Volume 6, Issue 1

46
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
Number of folds can be the same as the number of
instances but to obtain accuracy folds should be between
5 to 20, not more than that. If folds are set to 10, then 1
fold is taken for training and the remaining 9 are taken
for testing in the weka tool. Classiers, attributes, and
folds are predened tools in the weka tool. Machine
learning uses some specic mathematical methods to
train datasets which are known as classiers. In the
Weka tool, rstly an attribute is selected then folds
are set to 10, 12, or 8 (as per wish), then a classier
is selected, and nally after some preparation time,
we obtain the results for that particular attribute and
classier. One after one attribute is trained and tested
to observe which attributes obtain the highest accuracy
among all attributes.
3.1.2. MATLAB Python Approach
MATLAB is a high-level language and interactive
environment for numerical computation, visualization,
and programming which is especially useful for
medical images processing. Also, Python is an
interpreted high-level general-purpose programming
language. Its amazing libraries and tools help in
achieving the task of image processing very eciently.
In this paper, for Alzheimer’s disease diagnosis, we
have used MATLAB for preprocessing, segmentation,
and feature extraction, and used Python to apply SVM,
KNN, DT, RF and MLP classiers on the feature
matrix. In this approach, dataset analysis is done by
Preprocessing, Segmentation, Feature extraction, and
classifying the dataset attributes. A owchart of this
approach is shown in Figure 3.
For the Alzheimer’s disease study, we obtained
data from the Alzheimer’s Disease Neuroimaging
Initiative (ADNI) database (adni.loni.usc.edu). The
ADNI was launched in 2003 by the National Institute
on Aging (NIA), the Food and Drug Administration
(FDA), the National Institute of Biomedical Imaging
and Bioengineering (NIBIB), non-prot private
pharmaceutical companies, and other organizations,
with funding of $60 million for the ve-year private-
public partnership (62). In this study, 50 MRI images
data were collected from healthy people (Mean of
age ± STD = 77.92 ± 5.17 and Mean of weight ±
STD = 78.62 ± 21.32) and people with Alzheimer’s
disease (Mean of age ± STD = 74.4 ± 9.82 and Mean
of weight ± STD = 79.14 ± 12.26).
Figure 3. General steps to diagnosis of Alzheimer’s disease based on MRI images using MATLAB and Python
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

47
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
All the participants in this study were scanned
with GE Medical Systems or SIEMENS or Philips
Medical Systems MRI scanner with 1.5 or 3 Tesla
eld strength. The T1-weighted MRI scans were
captured with a coronal acquisition plane.
1. Preprocessing
In the analyzing process, at the rst, pre-processing
is performed to increase the quality of images. So
the median lter was used to remove the noise in the
images. The median lter is the ltering technique
used for noise removal from images and signals.
Median lter is very crucial in the image processing
eld as it is well known for the preservation of edges
during noise removal (63).
2. Segmentation
Image segmentation is the process of partitioning an
image into multiple segments (64). Image segmentation
is typically used to locate objects and extract regions
of interest in an image. In this paper, for the diagnosis
of Alzheimer’s disease, the lateral ventricles regions,
hippocampus, and some areas of brain tissue are
considered so that after their segmentation, features
can be extracted from these areas.
Table 2. Accuracy of Personality Disorder, Depression, Anxiety and Schizophrenia for dierent classiers
CCllaassssiiffiieerr
DDiissoorrddeerr
AAttttrriibbuutteess aanndd TThheeiirr AAccccuurraaccyy
SVM
Personality Disorder
Elapse - 99.99%
Gender - 58.86%
Score - 99.55%
Depression
Age - 37.30%
Married - 80.05%
Incoming Salary - 82.01%
Anxiety
Gender - 45.45%
Student - 75.75% (12 folds)
Age - 74.24%
Schizophrenia
Subject - 24.54%
Onset - 80.90%
Disorder - 70%
Logistic
Personality Disorder
Elapse - 99.99%
Gender - 63.70%
Score - 21.86%
Depression
Age - 37.57%
Married - 81.17%
Incoming Salary - 79.71%
Anxiety
Gender - 42.42%
Student - 60.60% (12 folds)
Age - 72.72%
Schizophrenia
Subject - 19.54% (12 folds)
Onset - 79.09% (12 folds)
Disorder- 66.18% (12 folds)
MLP
Personality Disorder
Elapse - 99.99%
Gender - 57.40%
Score - 21.06%
Depression
Age - 33.17%
Married - 81.24%
Incoming Salary - 77.32%
Anxiety
Gender - 42.42% (12 folds)
Student - 68.18%
Age - 74.24% (12 folds)
Schizophrenia
Subject - 16.36% (12 folds)
Onset - 71.36% (8 folds)
Disorder - 60.45%
December 2021, Volume 6, Issue 1

48
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
The lateral ventricles regions were extracted by
Otsu’s thresholding method and the hippocampus
region was extracted by rectangular drawing
method. Also, a skull stripping algorithm is used for
segmentation of brain tissues from the surrounding
region, and the Gray Matter (GM), White Matter
(WM), Cerebral Spinal Fluid (CSF) were extracted
using dierent methods of segmentation and
thresholding.
3. Feature extraction
After the segmentation step, the area of each
regions; lateral ventricles, hippocampus and brain
tissues was calculated as a feature (Since the images
are two-dimensional, the volume is equal to the
area). Also, according to the changes in the intensity
of the hippocampus, the statistical features such as
mean and standard deviation from this region as well
as texture features such as Gray Level Co-occurrence
Matrix (GLCM), which includes correlation,
contrast, and entropy, were extracted from this region
as features. From ADNI, the scores of persons in the
Mini-Mental State Examination (MMSE) and their
age were also obtained to form the feature matrix.
Therefore, a total of twelve features were extracted
for each individual.
4. Classication
After obtaining the feature matrix for all
individuals, we used the Training/Testing method
to separate 70% of the data for training algorithms
and 30% of the data for testing algorithms. Finally,
ve classications (KNN, SVM, DT, RF, and MLP)
were used to distinguish between healthy people and
people with Alzheimer’s. Then, Accuracy, sensitivity,
and specicity were used to evaluate each of the
classiers. The results for each of the classiers are
shown in Table 3.
4. Results and Discussion
The proposed system trains and tests the model
for classifying the data using certain classiers.
The application of all the classiers- SVM, KNN,
Logistic, MLP, DT, and RF- are shown in Figure 5,
KKNNNN
SSVVMM
DDTT
RRFF
MMLLPP
Accuracy
0.90 ± 0.08
0.94 ± 0.05
0.91 ± 0.03
0.94 ± 0.07
0.92 ± 0.06
Sensitivity
0.89 ± 0.10
0.94 ± 0.08
0.90 ± 0.11
0.96 ± 0.09
0.92 ± 0.13
Specificity
0.93 ± 0.15
0.93 ± 0.08
0.92 ± 0.09
0.93 ± 0.08
0.93 ± 0.08
Table 3. Results of average accuracy, sensitivity and specicity in 10 trials obtained from classiers for Alzheimer’s Disease (Rounded, and
Mean ± SD)
SSppeeaarrmmaann ccoorrrreellaattiioonn
PPeeaarrssoonn ccoorrrreellaattiioonn
Lateral ventricle (LV) size
0.419
0.338
Hippocampus (HP) size
-0.836
-0.611
Mean of intensity
-0.143
0.068
STD of intensity
0.387
0.276
Contrast of intensity
-0.090
-0.070
Correlation of intensity
-0.373
-0.411
Entropy of intensity
-0.050
0.030
White matter size
-0.164
-0.219
Gray matter size
-0.137
-0.224
Cerebral spinal fluid size
0.215
-0.088
MMSE
-0.850
-0.820
Age
-0.181
-0.223
Table 4. e Spearman correlation and Pearson correlation for each extracted features for AD diagnosis
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

49
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
Table 2, and Table 3.
The tables below contain attributes selected for
classication and their corresponding accuracies,
and in case of AD, their corresponding sensitivity and
specicity is also given. The attributes were selected
according to their eect on the data provided in the
dataset. For personality disorder, depression, anxiety,
and schizophrenia, three attributes have been selected
from the datasets. From the table given below we can
observe that for Personality Disorder, Depression,
Anxiety, and Schizophrenia, SVM performs the best
and MLP showed the least overall performance. For
AD, SVM and RF gave the best accuracy.
For the Alzheimer’s disease study, the Spearman
correlation, Pearson correlation and Mutual
Information were calculated to evaluate each of the
extracted features in the diagnosis of Alzheimer’s
disease. The results are shown in Table 4 and Figure 4.
Figure 4. Impact of each feature on Alzheimer's diagnosis based on Mutual Information
Figure 5. Results of each classier in each psychiatric disorder
December 2021, Volume 6, Issue 1

50
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
The above spider graph is plotted using data of
table 2 and table 3. It shows the average accuracy for
each disorder according to the classiers.
5. Conclusion
In this research paper, SVM was better in
comparison to the other two techniques. AI in
psychiatric disorders uses computerized techniques
as well as algorithms for diagnosis, prevention, and
treatment of mental disorders. Such techniques will
help society by diagnosing the disorder eectively
and nding out the proper medication and treatment.
Moreover, psychiatrists will be able to understand
and easily nd out the disorder.
In the Alzheimer’s disease study, based on
the Spearman correlation, Pearson correlation,
and Mutual information, the extracted features
were suitable features for Alzheimer’s diagnosis.
According to the study of papers, despite Alzheimer’s
disease, the lateral ventricles regions become larger
and the hippocampus region becomes smaller, which
in this study also follows these changes. Also, in the
classication step, we tested several classiers to
nd appropriate classiers. Overall, the proposed
model with the RF and SVM achieved the best
performance and the accuracy of these classiers
using the proposed method are 94%.
References
(1) Perez Arribas, et al. A signature-based machine
learning model for distinguishing bipolar disorder
and borderline personality disorder. Transl Psychiatry
8, 274 (2018). https://doi.org/10.1038/s41398-018-
0334-0
(2) U. Rajendra Acharya, et al. Automated EEG-based
Screening of Depression Using Deep Convolutional
Neural Network. Journal of Computational Science,
Elsevier, Volume 161, July 2018, https://doi.
org/10.1016/j.cmpb.2018.04.012
(3) Shrivastava, A. et al. A SVM-based classication
approach for obsessive compulsive disorder by
oxidative stress biomarkers, Journal of Computational
Science, Elsevier, Volume 36, September 2019,
101023, https://doi.org/10.1016/j.jocs.2019.07.010
(4) Saeedi, M., et al. Major depressive disorder
assessment via enhanced k-nearest neighbor method
and EEG signals. Phys Eng Sci Med 43, 1007–
1018 (2020). https://doi.org/10.1007/s13246-020-
00897-w
(5) Cremers, H., et al. Borderline personality disorder
classication based on brain network measures
during emotion regulation. Eur Arch Psychiatry
Clin Neurosci 271, 1169–1178 (2021). https://doi.
org/10.1007/s00406-020-01201-3
(6) Yang, X., Hu, X., Tang, W. et al. Multivariate
classication of drug-naive obsessive-compulsive
disorder patients and healthy controls by applying
an SVM to resting-state functional MRI data. BMC
Psychiatry 19, 210 (2019). https://doi.org/10.1186/
s12888-019-2184-6
(7) M. Khazbak, Z. Wael, Z. Ehab, et al. “MindTime:
Deep Learning Approach for Borderline Personality
Disorder Detection,” 2021 International Mobile,
Intelligent, and Ubiquitous Computing Conference
(MIUCC), 2021, pp. 337-344, doi: 10.1109/
MIUCC52538.2021.9447620.
(8) Bracher-Smith, M., Crawford, et al. Machine
learning for genetic prediction of psychiatric
disorders: a systematic review. Mol Psychiatry 26,
70–79 (2021). https://doi.org/10.1038/s41380-020-
0825-2
(9) A. Saidi, S. B. Othman et al. “Hybrid CNN-
SVM classier for ecient depression detection
system,” 2020 4th International Conference on
Advanced Systems and Emergent Technologies
(IC_ASET), 2020, pp. 229-234, doi: 10.1109/IC_
ASET49463.2020.9318302.
(10) Anu Priya et al. “Predicting Anxiety, Depression
and Stress in Modern Life using Machine Learning
Algorithms” Procedia Computer Science, Elsevier,
Volume 167, 2020, https://doi.org/10.1016/j.
procs.2020.03.442
(11) Pan, Zhongde et al. “Detecting Manic State of
Bipolar Disorder Based on Support Vector Machine
and Gaussian Mixture Model Using Spontaneous
Speech.” Psychiatry investigation vol. 15,7 (2018):
695-700. doi:10.30773/pi.2017.12.15
(12) Suyash Dabhane et al. “Depression Detection on
Social Media using Machine Learning Techniques”,
International Research Journal of Engineering and
Technology (IRJET) ,Volume: 07 Issue: 11, 2020.
(13) M. R. Islam, A. R. M. Kamal, et al. “Detecting
Depression Using K-Nearest Neighbors (KNN)
Classication Technique,” 2018 International
Conference on Computer, Communication, Chemical,
Material and Electronic Engineering (IC4ME2),
2018, pp. 1-4, doi: 10.1109/IC4ME2.2018.8465641.
(14) Shuojia Wang “Using Electronic Health Records
and Machine Learning to Predict Postpartum
Depression (PPD)”Volume 264: MEDINFO
2019: Health and Wellbeing e-Networks for All,
DOI:10.3233/SHTI190351
(15) A. Al-Ezzi, N. Yahya, N. Kamel, et al. “Severity
Assessment of Social Anxiety Disorder Using Deep
Learning Models on Brain Eective Connectivity,”
in IEEE Access, vol. 9, pp. 86899-86913, 2021, doi:
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

51
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
10.1109/ACCESS.2021.3089358.
(16) Arkaprabha Sau, Ishita bhakta,”Screening
of anxiety and depression among seafarers using
machine learning technology”, Informatics in
Medicine Unlocked, Elsevier, Volume 16, 2019,
https://doi.org/10.1016/j.imu.2019.100228
(17) Neesha Jothi et al,”Predicting generalized
anxiety disorder among women using Shapley
value”,Journal of Infection and Public Health,
Elsevier, Volume 14, Issue 1, January 2021, https://
doi.org/10.1016/j.jiph.2020.02.042
(18) Miseon Shim et al,”Machine-learning-based
classication between post-traumatic stress disorder
and major depressive disorder using P300 features”,
NeuroImage: Clinical, Volume 24, 2019, https://doi.
org/10.1016/j.nicl.2019.102001
(19) Xing M, Fitzgerald JM and Klumpp H (2020)
Classication of Social Anxiety Disorder With
Support Vector Machine Analysis Using Neural
Correlates of Social Signals of Threat. Front.
Psychiatry 11:144. doi: 10.3389/fpsyt.2020.00144
(20) H. Alharthi, “Predicting the level of generalized
anxiety disorder of the coronavirus pandemic among
college age students using articial intelligence
technology,” 2020 19th International Symposium on
Distributed Computing and Applications for Business
Engineering and Science (DCABES), 2020, pp. 218-
221, doi: 10.1109/DCABES50732.2020.00064.
(21) Kyoung-SaeNa “Machine learning-based
discrimination of panic disorder from other
anxiety disorders”, Journal of Aective Disorders,
Elsevier,Volume 278, 1 January 2021, https://doi.
org/10.1016/j.jad.2020.09.027
(22) Renzhou Gui : “The Impact of Emotional Music
on Active ROI in Patients with Depression Based on
Deep Learning: A Task-State fMRI Study”: Hindawi,
Computational Intelligence and Neuroscience,
Volume 2019, Article ID 5850830, https://doi.
org/10.1155/2019/5850830
(23) Wen Dong, Wei Zhenhao, et al.”Deep Learning
Methods to Process fMRI Data and Their Application
in the Diagnosis of Cognitive Impairment: A Brief
Overview and Our Opinion” JOURNAL=Frontiers
in Neuroinformatics, VOLUME=12, 2018,
DOI=10.3389/fninf.2018.00023
(24) Liu M, Cheng D and Yan W (2018) Classication
of Alzheimer’s Disease by Combination of
Convolutional and Recurrent Neural Networks Using
FDG-PET Images. Front. Neuroinform. 12:35. doi:
10.3389/fninf.2018.00035
(25) A. Sau, I. Bhakta, Predicting anxiety and
depression in elderly patients using machine learning
technology, Healthcare Technology Letters, 4 (6)
(2017 Nov 10), pp. 238-243
(26) R. S. McGinnis et al., “Rapid Anxiety and
Depression Diagnosis in Young Children Enabled
by Wearable Sensors and Machine Learning,” 2018
40th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society
(EMBC), 2018, pp. 3983-3986, doi: 10.1109/
EMBC.2018.8513327.
(27) E. W. McGinnis et al., “Giving Voice to
Vulnerable Children: Machine Learning Analysis
of Speech Detects Anxiety and Depression in Early
Childhood,” in IEEE Journal of Biomedical and
Health Informatics, vol. 23, no. 6, pp. 2294-2301,
Nov. 2019, doi: 10.1109/JBHI.2019.2913590.
(28) Nemesure, M.D., Heinz, M.V., Huang, R. et
al. Predictive modeling of depression and anxiety
using electronic health records and a novel machine
learning approach with articial intelligence. Sci Rep
11, 1980 (2021). https://doi.org/10.1038/s41598-
021-81368-4
(29) Richter, T., Fishbain, B., Markus, A. et al. Using
machine learning-based analysis for behavioral
dierentiation between anxiety and depression.
Sci Rep 10, 16381 (2020). https://doi.org/10.1038/
s41598-020-72289-9
(30) ZhiHong Chen, Tao Yan, ErLei Wang, Hong
Jiang, YiQian Tang, Xi Yu, Jian Zhang, Chang Liu,
“Detecting Abnormal Brain Regions in Schizophrenia
Using Structural MRI via Machine Learning”,
Computational Intelligence and Neuroscience, vol.
20200, Article ID 6405930, 13 pages, 2020. https://
doi.org/10.1155/2020/6405930
(31) Calhas et al.”On the use of pairwise distance
learning for brain signal classication with limited
observations”, Articial Intelligence in Medicine,
Elsevier, Volume 105, May 2020, https://doi.
org/10.1016/j.artmed.2020.101852
(32) Tharindu Fernando, et al. “Neural memory
plasticity for medical anomaly detection”,
Neural Networks, Volume 127, 2020, https://doi.
org/10.1016/j.neunet.2020.04.011.
(33) Guo Y, Qiu J, Lu W. Support Vector Machine-
Based Schizophrenia Classication Using
Morphological Information from Amygdaloid and
Hippocampal Subregions. Brain Sciences. 2020;
10(8):562. https://doi.org/10.3390/brainsci10080562
(34) C. -R. Phang, F. Noman, et al., “A Multi-Domain
Connectome Convolutional Neural Network for
Identifying Schizophrenia From EEG Connectivity
Patterns,” in IEEE Journal of Biomedical and Health
Informatics, vol. 24, no. 5, pp. 1333-1343, May
2020, doi: 10.1109/JBHI.2019.2941222.
(35) Oh J, Oh B-L, Lee K-U, Chae J-H and Yun K
(2020) Identifying Schizophrenia Using Structural
MRI With a Deep Learning Algorithm. Front.
December 2021, Volume 6, Issue 1

52
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
Psychiatry 11:16. doi: 10.3389/fpsyt.2020.00016
(36) T. Matsubara, T. Tashiro and K. Uehara, “Deep
Neural Generative Model of Functional MRI Images
for Psychiatric Disorder Diagnosis,” in IEEE
Transactions on Biomedical Engineering, vol. 66,
no. 10, pp. 2768-2779, Oct. 2019, doi: 10.1109/
TBME.2019.2895663.
(37) Alexandra Talpalaru, et al. Identifying
schizophrenia subgroups using clustering and
supervised learning, Schizophrenia Research,
Volume 214, 2019, https://doi.org/10.1016/j.
schres.2019.05.044.
(38) Sugai Liang, et al., Classication of First-
Episode Schizophrenia Using Multimodal Brain
Features: A Combined Structural and Diusion
Imaging Study, Schizophrenia Bulletin, Volume
45, Issue 3, May 2019, Pages 591–599, https://doi.
org/10.1093/schbul/sby091
(39) Chatterjee, Indranath et al. “Identication of
brain regions associated with working memory
decit in schizophrenia.” F1000Research vol. 8 124.
30 Jan. 2019, doi:10.12688/f1000research.17731.1
(40) Kalmady, S.V., Greiner, R., Agrawal, R. et al.
Towards articial intelligence in mental health by
improving schizophrenia prediction with multiple
brain parcellation ensemble-learning. npj Schizophr
5, 2 (2019). https://doi.org/10.1038/s41537-018-
0070-8
(41) Qureshi et al.”3D-CNN based discrimination
of Schizophrenia using resting-state fMRI” ,
ELSEVIER, Articial Intelligence in Medicine,
Volume 98, July 2019, https://doi.org/10.1016/j.
artmed.2019.06.003
(42) Wang Yu, Zhang Na, Yan Fengxia, Gao Yanping.
Magnetic resonance imaging study of gray matter
in schizophrenia based on XGBoost. Journal of
Integrative Neuroscience, 2018, 17(4): 331-336.
(43) Manohar, L., Ganesan, K. Diagnosis of
Schizophrenia Disorder in MR Brain Images Using
Multi-objective BPSO Based Feature Selection with
Fuzzy SVM. J. Med. Biol. Eng. 38, 917–932 (2018).
https://doi.org/10.1007/s40846-017-0355-9
(44) Latha, M., Kavitha, G. Segmentation and
texture analysis of structural biomarkers using
neighborhood-clustering-based level set in MRI of
the schizophrenic brain. Magn Reson Mater Phy 31,
483–499 (2018). https://doi.org/10.1007/s10334-
018-0674-z
(45) A. de Pierrefeu et al., “Interpretable and stable
prediction of schizophrenia on a large multisite
dataset using machine learning with structured
sparsity,” 2018 International Workshop on Pattern
Recognition in Neuroimaging (PRNI), 2018, pp. 1-4,
doi: 10.1109/PRNI.2018.8423946.
(46) Xiao, Y., et al., Support vector machine-
based classication of rst episode drug-naïve
schizophrenia patients and healthy controls using
structu..., Schizophr. Res. (2017), https://doi.
org/10.1016/j.schres.2017.11.037
(47) Lu, Xiaobing et al. “Discriminative analysis
of schizophrenia using support vector machine
and recursive feature elimination on structural
MRI images.” Medicine vol. 95,30 (2016): e3973.
doi:10.1097/MD.0000000000003973
(48) El-Sappagh, S., Alonso, J.M., Islam, S.M.R. et
al. A multilayer multimodal detection and prediction
model based on explainable articial intelligence
for Alzheimer’s disease. Sci Rep 11, 2660 (2021).
https://doi.org/10.1038/s41598-021-82098-3
(49) Umeda-Kameyama Y, Kameyama M, Tanaka
T, et al. Screening of Alzheimer’s disease by facial
complexion using articial intelligence. Aging
(Albany NY). 2021;13(2):1765-1772. doi:10.18632/
aging.202545
(50) Li, V.O.K., Lam, J.C.K., Han, Y. et al. Editorial:
Designing a Protocol Adopting an Articial
Intelligence (AI)–Driven Approach for Early
Diagnosis of Late-Onset Alzheimer’s Disease. J
Mol Neurosci 71, 1329–1337 (2021). https://doi.
org/10.1007/s12031-021-01865-z
(51) Tsuji, S., Hase, T., Yachie-Kinoshita, A. et
al. Articial intelligence-based computational
framework for drug-target prioritization and inference
of novel repositionable drugs for Alzheimer’s
disease. Alz Res Therapy 13, 92 (2021). https://doi.
org/10.1186/s13195-021-00826-3
(52) Bahado-Singh RO, Vishweswaraiah S, et
al. (2021) Articial intelligence and leukocyte
epigenomics: Evaluation and prediction of late-onset
Alzheimer’s disease. PLoS ONE 16(3): e0248375.
https://doi.org/10.1371/journal.pone.0248375
(53) Ludwig et al., “Machine Learning to Detect
Alzheimer’s Disease from Circulating Non-coding
RNAs”, ELSEVIER, Genomics, Proteomics and
Bioinformatics, Volume 17, Issue 4, August 2019,
https://doi.org/10.1016/j.gpb.2019.09.004
(54) B. Richhariya, M. Tanveer, … A. R.-… S. P.
and, and undened 2020, “Diagnosis of Alzheimer’s
disease using universum support vector machine
based recursive feature elimination (USVM-RFE),”
Elsevier.
(55) R. Khan, M. Tanveer, … R. P.-E., and undened
2020, “A novel method for the classication of
Alzheimer’s disease from normal controls using
magnetic resonance imaging,” Wiley Online Library,
(56) F. Zhang et al., “Voxel-based morphometry:
improving the diagnosis of Alzheimer’s disease
based on an extreme learning machine method from
December 2021, Volume 6, Issue 1
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies

53
JAMSAT
Journal of Advanced Medical Sciences and Applied Technologies
Bhavsar Y, et al. Articial intelligence and stochastic process-based analysis of human psychiatric
disorders. JAMSAT. 2021; 6(1): 33-53..
the ADNI cohort,” Elsevier.
(57) J. Islam, Y. Z.-B. informatics, and undened
2018, “Brain MRI analysis for Alzheimer’s disease
diagnosis using an ensemble system of deep
convolutional neural networks,” Springer.
(58) “30th European Symposium on Computer Aided
Chemical Engineering,” Accessed: Sep. 09, 2021.
(59) M.-S. Mushtaq and A. Mellouk, “Methodologies
for Subjective Video Streaming QoE Assessment,”
Quality of Experience Paradigm in Multimedia
Services, pp. 27–57, 2017, doi: 10.1016/B978-1-
78548-109-3.50002-3.
(60) R. U. Khan, M. Tanveer, and R. B. Pachori, “A
novel method for the classication of Alzheimer’s
disease from normal controls using magnetic
resonance imaging,” Expert Systems, vol. 38, no. 1,
Jan. 2021, doi: 10.1111/EXSY.12566.
(61) K. Sagayam, P. Bruntha, M. Sridevi, … M.
S.-… V. P. for, and undened 2021, “A cognitive
perception on content-based image retrieval using
an advanced soft computing paradigm,” Elsevier,
Accessed: Sep. 07, 2021.
(62) Y. Tan, “GPU-Based Parallel Implementation
of Swarm Intelligence Algorithms,” GPU-Based
Parallel Implementation of Swarm Intelligence
Algorithms, pp. 1–236, Apr. 2016, doi: 10.1016/
C2015-0-02468-6.
December 2021, Volume 6, Issue 1