Mental Health Prediction for Aged Person in Brain

MRI Images using Deep Learning

Capstone Project(CSE1904)

Review-III

School: SCOPE
Guide: Prof. Rajkumar R

Team Members:
1.Aishna Gupta - 19BCB0101
2.Sanjeev - 19BCE0181
3.Sudhanshu Gupta - 19BCE0188
CONTENTS:
1. Abstract
2. Literature survey
3. Aim of project
4. Objective of project
5. Motivation behind
6. System Architecture diagram
7. Module Description
8. Methodology
9. Conclusion and future scope
10. References
 Neuroimaging-based approaches have been extensively applied
to study brain disease in recent years and have deepened our
understanding of both cognitively healthy and disordered brain
structure and function. The individual prediction and
characterization of patients with psychiatric problems have
shown promising results regarding recent breakthroughs in
machine-learning approaches. To evaluate patients with

Abstract heterogeneous brain disorders like schizophrenia and bipolar

disorder, studies have used features from a variety of
neuroimaging modalities, such as structural, functional, and
diffusion magnetic resonance imaging data, as well as jointly
estimated features from multiple modalities. The proposed
method is used to predict whether the patient has any brain
disease or not along with illness types like schizophrenia and
bipolar disorder, using deep learning models. The performance
metrics like accuracy are used to compare which model is best.
AIM

The aim of the project is to develop an accurate and reliable deep learning-based
approach for the individual prediction and characterization of patients with psychiatric
disorders, specifically schizophrenia and bipolar disorder, using neuroimaging data. The
use of machine learning approaches can improve the diagnosis and treatment of
psychiatric disorders, as it allows for the identification of unique patterns and biomarkers
in brain structure and function that are difficult to detect through traditional clinical
methods.
Objective

Early detection of mental health problems is an essential step in better understanding

mental illness and ensuring better patient care. In contrast to the diagnosis of other
chronic diseases, which rely on laboratory tests and measurements, mental illness is
usually diagnosed with special questionnaires designed to identify specific patterns of
emotional and social interactions. Diagnosis is based on individual self-reporting. Due to
the increasing availability of data on an individual's mental health, artificial intelligence
and machine learning techniques are being used to improve our understanding of mental
health. We are obligated to help providers improve their clinical situation. support
decision-making.
Motivation
The motivation behind this project stems from the urgent need to tackle the widespread
problem of mental health disorders, especially among older individuals. With millions of
people globally affected by these conditions, it is crucial to identify them early and provide
proper care. The project aims to use advanced technology, specifically deep learning and
imaging techniques, to create a reliable and affordable method for detecting brain diseases.
By achieving this goal, the project can significantly improve the lives of those affected,
offering timely interventions, better decision-making by healthcare providers, and
ultimately, a higher quality of life. It showcases the commitment to using technology to
address real-world health challenges and bring about positive change for people struggling
with mental illnesses.
Literature Survey

Identifying Schizophrenia Using Structural MRI with a Deep Learning Algorithm

[2020]
Skills Identified Gaps identified

● The ability of deep learning algorithms, ● While our project focuses on predicting the
specifically convolutional neural networks occurrence of schizophrenia and bipolar disorder
(CNNs), to accurately detect schizophrenia from in aged individuals using MRI images, the
structural MRI data. literature review only focuses on detecting
● The potential of deep learning algorithms to schizophrenia from structural MRI data
identify relevant structural features of ● Our project utilizes two deep learning models,
schizophrenia, such as the right temporal and VGG16 and Convolutional Neural Networks
parietal areas of the brain. (CNN), while the literature review only focuses
on the use of CNNs for detecting schizophrenia
from structural MRI data.
Literature Survey

Use of machine learning approach to predict depression in elderly in China: A

longitudinal study [2020]

Skills Identified Gaps Identified

● Knowledge of machine learning algorithms and ● Limited generalizability of findings due to small
their application in medical research, sample sizes and lack of diversity in study
particularly in predicting or diagnosing mental populations.
health conditions. ● Need for more rigorous evaluation and
● Ability to collect and analyze longitudinal data replication of findings to ensure that machine
from large, nationally representative surveys. learning models can be integrated into clinical
practice and improve patient outcomes.
● Lack of attention to ethical considerations, such
as issues related to data privacy and potential
bias in machine learning models.
Literature Survey

Diagnosis of Schizophrenia based on deep learning using fMRI [2021]

Skills Identified Gaps Identified

● This include knowledge of schizophrenia as a ● limited use of only one type of MRI modality,
brain disorder, familiarity with MRI and fMRI and the lack of external validation of the results.
imaging techniques ● Future studies should address these limitations
● understanding of machine learning algorithms, to ensure the reproducibility and generalizability
particularly convolutional neural networks, and of the findings.
expertise in data analysis and interpretation.
● The study also demonstrates skills in
experimental design and use of statistical
analysis.
Literature Survey

Prediction of Mental Health Among University Students. [2021]

Skills Identified Gaps Identified

● Data analysis and preprocessing ● The study had a relatively small sample size,
● Machine learning algorithm selection and which may limit the generalizability of the
implementation findings to larger populations.
● Model evaluation using various metrics ● The data used in the study was self-reported,
● Interpretation of results and identification of which may introduce biases and inaccuracies in
potential implications for mental health the responses.
interventions and services
S
Y
S
T
E
M

A
R
C
H
I
T
E
C
T
U
R
E
System Model Description
➔ Data Collection
The dataset will be collected from various platforms like Center for Biomedical
Research Excellence (COBRE), Northwestern University Schizophrenia Data and
Software Tool (NUSDAST), domain experts, literature survey and other accessible
platforms.

➔ Data Augmentation
We enrich the training data to achieve a balanced distribution among the classes
because the image classes are severely unbalanced. To build a new augmented
dataset, we rotate and mirror the photos.
➔ Splitting of data
Data is standardized for the model's training and testing once it has been cleaned.
After the data is split, we train the algorithm on the training data set while putting
the test data aside. Based on the logic, methods, and values of the feature in the
training data, this training process will generate the training model. The primary
goal of feature extraction is to scale all values uniformly.
● Training set: A data scientist utilizes a training set to train a model and
determine its ideal parameters- parameters it must learn from the data.
● Test set: For an evaluation of the trained model's generalizability, a test set is
required. The latter refers to a model's capacity to find patterns in fresh,
unexplored data after being trained on training data. The inability for
generalization we described earlier, known as model overfitting, must be
avoided by using distinct subsets for training and testing.
➔ Classification
Deep Learning Technique is used to predict mental illness using
classification and ensemble techniques on brain MRI datasets. The
objective is to analyze the performance of these methods and find
accuracy, and identify the responsible/important features which play
a major role in prediction. So for comparing the accuracy and other
features we would be using two algorithms:
1. CNN(Convolutional Neural Network)
2. VGG16 model
Comparison
Once we have successfully implemented both the algorithms we will compare
the results using the accuracy and confusion matrix.

1. Accuracy: The accuracy metric is one of the simplest Classification

metrics to implement, and it can be determined as the number of correct
predictions to the total number of predictions.
2. Confusion matrix: It is an NxN matrix that describes the overall
performance of a model when used on some dataset, where N is the
number of class labels in the classification problem. It clearly tells the
correct and incorrect prediction in the form of True Positives, False
Positives, True Negatives, False Negatives.
 Methodology Adapted

CNN VGG16
Within Deep Learning, a Convolutional VGG16 is a convolution neural
Neural Network or CNN is a type of net (CNN ) architecture which
artificial neural network, which is was used to win
widely used for image/object ILSVR(Imagenet) competition in
recognition and classification. Deep 2014. It is considered to be one of
Learning thus recognizes objects in an the excellent vision model
image by using a CNN. architecture
CNN
CNN(Convolution Neural Network)
The implementation of the CNN algorithm is done in following steps:

1. Classifier Training: A classifier is a function that takes features as input and generates a
class label prediction. Already Neuroimaging studies have applied various classifiers for
mental illness prediction, but there are dimensional issues which should be taken into
account before applying such classification algorithms.
2. Convolutional layer: This layer extracts the features from the image which is applied as
input. The neurons convolve the input image and produce a feature map in the output image
and this output image from this layer is fed as an input to the next convolutional layer.
3. Pooling layer: This layer is used to decrease the dimensions of the feature map still
maintaining all the important features. This layer is usually placed between two
convolutional layers.
4. Relu layer: A rectified linear unit (ReLU) is an activation function which interprets the positive
part of its argument.

5. Fully Connected layer: FLC means that each filter in the previous layer is connected to each
filter in the next layer. This is used to classify the input image based on the training dataset into
various classes.

It has 4 phases:
● Model construction: Model construction depends on machine learning algorithms. In this
projects case, it was Convolution Neural Networks.
● Model training: Here, the model is trained using training data and expected output for this
data. Once the model has been trained it is possible to carry out model testing.
● Model testing: During this phase a second set of data is loaded. This data set has never been
seen by the model and therefore it’s true accuracy will be verified.
● Model evaluation : After the model training is complete, the saved model can be used in the
real world. The name of this phase is model evaluation.
VGG16
Algorithm
The ImageNet dataset contains images of fixed size of 224*224 and have RGB channels. This model process the
input image and outputs the a vector of 1000 values. This vector represents the classification probability for the
corresponding class. To make sure these probabilities are added, we use softmax function.

Softmax Function uses the 5 most probable candidates into the vector.and our ground truth vector which then uses
the complex mathematical function to derive the results.

Architecture: The input to the network is an image of dimensions (48, 48, 3). The first two layers have 64 channels
of 3*3 filter size and same padding, followed by two layers of 256 filter size and padding size (3, 3). After that,
there are 2 sets of 3 convolution layers and a max pool layer. The stack of convolution and max-pooling layers
produces a (7, 7, 512) feature map, flatten it to a (1, 25088) vector, and pass it to a softmax layer to normalize the
classification vector. All the hidden layers use ReLU as their activation function, which is more computationally
efficient and decreases the likelihood of vanishing gradient problems.
Conclusion

In conclusion, this project demonstrates the potential of deep learning techniques,

specifically Convolutional Neural Networks (CNN) and the VGG16 model, in predicting
brain diseases, such as schizophrenia and bipolar disorder, in aged individuals based on
MRI images. The results indicate that the VGG16 model outperformed the CNN model,
achieving an impressive accuracy of 100% compared to 79.63%. It is essential to
acknowledge the limitations of the project, such as the small sample size of MRI images
utilized, which may impact the generalizability of the findings. Future studies should
focus on expanding the sample size and including more diverse populations to enhance
the applicability and validity of the results.
REFERENCES
1. Oh J, Oh BL, Lee KU, Chae JH, Yun K. Identifying Schizophrenia Using Structural MRI With a Deep Learning Algorithm. Front Psychiatry.
2020;11:16. doi:10.3389/fpsyt.2020.00016
2. Su, Dai & Zhang, Xingyu & He, Kevin & Chen, Yingchun. (2020). Use of machine learning approach to predict depression in the elderly in China: A
longitudinal study. Journal of Affective Disorders. 282. 10.1016/j.jad.2020.12.160.
3. Nunes A, Schnack HG, Ching CRK, et al. Using structural MRI to identify bipolar disorders - 13 site machine learning study in 3020 individuals from
the ENIGMA Bipolar Disorders Working Group. Mol Psychiatry. 2020;25(9):2130-2143. doi:10.1038/s41380-018-0228-9
4. Sharma, S., Guleria, K., Tiwari, S., & Kumar, S. (2022). A deep learning based convolutional neural network model with VGG16 feature extractor for
the detection of Alzheimer Disease using MRI scans. Measurement: Sensors, 24, 100506. DOI: 10.1016/j.measen.2022.100506
5. Wong K, Zheng J, Wei X, Wang J, Lin H, Pan H, Shi Y. Diagnosis of Schizophrenia Based on Deep Learning Using fMRI. 2021;2021:8437260.
6. Sairam U, Voruganti S. Mental Health Prediction Using Deep Learning. Int J Res Appl Sci Eng Technol. 2022;2(2). DOI:10.22214/ijraset.2022.40371
7. Nguyen, Le & Do, Thi Thu Ha. (2019). Artificial Intelligence in Healthcare: A New Technology Benefit for Both Patients and Doctors. 1-15.
10.23919/PICMET.2019.8893884.
8. Yadav, Samir & Jadhav, Shivajirao. (2019). Deep convolutional neural network based medical image classification for disease diagnosis. Journal of Big
Data. 6. 10.1186/s40537-019-0276-2.
9. Sahlan, Fadhluddin & Mohammad Nizam, Faris Hamidi & Misrat, Muhammad & Zamzuri, Muhammad & Wani, Sharyar & Gulzar, Yonis. (2021).
Prediction of Mental Health Among University Students. International Journal on Perceptive and Cognitive Computing. 7. 85-91.
10. Vandana, Marriwala N, Chaudhary D. A hybrid model for depression detection using deep learning. Meas Sensors. 2023;25:100587.
doi:10.1016/j.measen.2022.100587