Predictive Analytics in Healthcare System using Deep Learning

Approach

Salma Lahlali and Abdelkarim Ammoumou

Industrial Engineering, Signal Processing and Logistic Laboratory, FSAC, Hassan 2nd University, Casablanca, Morocco

Keywords: Deep Learning; Healthcare; Internet of things

Abstract: During centuries, and since the appearance of science, the world of healthcare has known a noticeable

positive progress. Not only on the researches and inventions term, but also on the intention’s term. The first

aim was to save people’s lives and heal patients, from the illnesses which was known back then. After a

period, and thanks to the improvement that was known in the field of both science & research, scientists

level up to another aim which is preventing. This means, that they were looking for solutions by which

doctors will be able to predict a possible disease infection, thus the possibility to prevent an exposure to a

disease was being possible. Therefore, saving people’s lives had a new road which needed to be developed

with the cooperation of other domains, so a system can be put into action to realize this new mission. The

purpose of this article, is to present a system which can predict a possible illness as well as suggest the

adequate treatment for the patient’s case. This includes discussing the system’s main objectives and

characteristics, in addition to describing its architecture main layers.

1 INTRODUCTION

Whith the evolvement of the information

technology, many domains has been interested in

using the knowledge provided by the IT to improve

their services.

As shown in the Figure 1 and based on Mc

Kinsey’s report about the economic impact of IoT

by 2025, the impact will reach the 6.2$ trillion.

So as we can see, on the presentation the higher

percentage goes to the healthcare domain, which

means that 43% of this domain is impacted by IoT.

In the second rank, comes the industry with 34%,

other domains such as transportation, agriculture,

urban infrastructure, security and retail come next

with 15% and finally the 7% left is reserved for Iot

itself.

Figure 1: Percentage of impacted domain by IoT in 2025

As the world is progressing on every single way,

the information has been the center of the interest of

many sciences.

Therefore, many techniques have been

developed. Such as, deep learning, which is known

also as deep structured learning/ hierarchical

learning.

It’s a sort of machine learning methods that is

based on learning data representations.

So, we have now two huge concepts: Information

Technology & Deep Learning, if we succeeded to

combine them, this will for sure make an impressive

Lahlali, S. and Ammoumou, A.

Predictive Analytics in Healthcare System using Deep Learning Approach.

DOI: 10.5220/0009776500110015

In Proceedings of the 1st International Conference of Computer Science and Renewable Energies (ICCSRE 2018), pages 11-15

ISBN: 978-989-758-431-2

impact on getting the right results in the healthcare

field.

It will increasingly improve the healthcare for

both individuals and communities, that’s because the

deep models enable the discovery of high-level

features, improve performances and provide

additional understanding.

So we will be presenting an overview of deep

learning, the most trending architectures and

frameworks that have been introduced in recent

years. Then, we will proceed to describe our

contribution through presenting an intelligent system

which was conceived to provide predictive analytics

in healthcare, using deep learning approach. After

that, we will denote this system’s ability to predict

the probability for a patient to develop a specific

disease as well as to offer a personalized treatment.

2 STATE OF ART: DEEP

LEARNING

Deep Learning architectures have gained more

attention in recent years compared to the other

traditional machine learning approaches. Deep

learning refers to a set of machine learning

techniques that learn multiple levels of

representations.

Its architectures consist of multiple processing

layers: the input layer, several hidden layers, and an

output layer. Each one contains neurons.

In figure 2, the input image is convolved with

three trainable filters and biases to produce three

feature maps at the C1 level. Each group of four

pixels in the feature maps are added, weighted,

combined with a bias, and passed through a sigmoid

function to produce the three feature maps at S2.

These are again filtered to produce the C3 level. The

hierarchy then produces S4 in the same way the S2

was produced. Finally, these pixel values are

rasterized and presented as a single vector input to

the “conventional” neural network at the output.

Figure 2: Conceptual example of convolutional neural

network

The table 1 below presents a brief overview of

the most trending architectures that have been

introduced in deep learning in recent years:

Table1 : The most trending architectures introduced in

Deep Learning last years

Model

earning

odel

Typical

input

data

Description

CNN

Supervised

 Convolution

layer take biggest part

of computations

▪ Every hidden

convolutional filter

transforms its input to

a 3D output volume

of neuron activations

▪ Inspired by the

neurobiological model

of the visual cortex

Application example:

Alzheimer diagnosis

RNN

Supervised

Serial,

time series

▪ Useful in IoT

applications with

time-dependent data

▪ All the layers

share the same

weights

▪

Application example:

Human behavior

monitoring

ICCSRE 2018 - International Conference of Computer Science and Renewable Energies

The trending of Deep Learning architectures

imposes the introduction of Deep Learning

frameworks. They help easily create and test various

deep architectures. In this section, we list some of

these frameworks:

● H2O: Use Java as a core language and can

interface with R, Python, Scala, Java, JSON, and

CoffeeScript/JavaScript. H2O is used for critical

applications like predictive maintenance and

operational intelligence. H2O includes many

common machine learning algorithms, such as

generalized linear modeling.12

● Tensorflow: It is an open-source software

library Developed with C++. Tensorflow’s platform

includes interfaces for Python, Java, C, and C++.

Tensorflow supports both large-scale training and

inference. It can be a support to visualize networks.

At the same time, it is flexible enough to support

experimentation and research into new machine

learning models and system-level optimizations. (9)

● Caffee: The framework is a C++ library

with Python and MATLAB bindings for training and

deploying general purpose convolutional neural

networks and other deep models efficiently on

commodity architectures. It powers ongoing

research projects, large-scale industrial applications,

and startup prototypes in vision, speech, and

multimedia. 11

● Theano: is a free Python symbolic

manipulation library. It has specifically been utilized

for the gradient-based methods such as deep learning

that require repeated computation of the tensor-

based mathematical expressions. It offers for

implementing standard and non-standard deep

architectures. 10

3 SYSTEM ARCHITECTURE

Because prevention is better than cure, the main

purpose of the system is to predict complexity and

pathologies that any person can have in the near

future. More importantly, the system will not only

predict the diseases, but also propose the suitable

treatment.

3.1 System Characteristics

3.1.1 Description

The main actors of the system are the patient and

the doctor. The interaction of those two will be as

follows:

● The doctor can have access to the system

in order to consult the patient case permanently, or if

a patient asks for help. The doctor must also

intervene if he is notified for a probable disease for a

patient.

● The patient interacts with the system to

solicit a doctor’s help or to follow the doctor’s

guidelines. The patient must be connected to IoT to

collect information about the environment around it.

Example: EHS, Electronic Health Records

LSTM

Supervised

Serial,

time

series,

long

time

depende

nt data

▪ Modeling the hidden

state with cells that decide

what to keep in memory

given the previous state,

the current memory and

the input value.

▪ Good performance

with data of long time lag

RBM

pervised Unsupervise

Various ▪ A variant of

Boltzmann machines,

which is a type of

stochastic neural network

▪ Useful if it is

required to model

probabilistic relationships

between variables 777

Application example:

Human activity

recognition

DBN

pervised, Unsupervise

Various ▪ A special BM where

the hidden units are

organized in a deep

layered manner, only

adjacent layers are

connected, and there are

no visible–visible or

hidden–hidden

connections within the

same layer.

Application example:

cancer diagnosis

nsupervise

Various ▪ Trained to minimize

the reconstruction error

▪ Mostly used for

representation learning

▪ Same number of

input and output

Application: 3D brain

reconstruction

Predictive Analytics in Healthcare System using Deep Learning Approach

3.1.2 Characteristics

The system will:

• Be a source of almost everything that doctors

will need about the patients.

• Be hosted in cloud.

• Include security measure to ensure that the

access to the patient information will be limited to

only the person whom the patient grants overt

access.

The analyst will be in real time.

The figure 3 illustrates the interactions between

the doctor, the patient and the healthcare system.

Figure 3: Schema of communication healthcare system

3.2 System Architecture

To describe the system’s architecture, we should

introduce the two layers of the system which are:

- Data collection & Treatment of Data layer.

- Predictive Analysis layer.

The figure 4 represents the layers of our

predictive analysis system healthcare.

3.2.1 Data Collection & Treatment of Data

A formal data collection process is necessary. It

ensures that the data is defined and accurate. The

healthcare data coming from EHS, biomedical

database and public health are one of the used IoT

equipment. They have been enhanced not only on

the availability and traceability but also on the

liquidity of data.

Once the data is collected, the phase of treatment

starts. This process is as important as the data

collection process. To realize this huge task, we

resort Spark. Apache Spark is an open-source

platform for large-scale data processing that is well-

suited for iterative machine learning tasks. It can be

interactively used to quickly process and query big

datasets.

3.2.2 Predictive Analysis Module

As mentioned above, our system’s main goal is

prediction. So, the predictive analysis module is the

master of the system. At this level, the system

analyzes the current state in addition to the medical

history to make predictions about possible future

illnesses.

To produce a tangible product that provides right

decision with accessible and useful information, we

need to choose the correct architecture and the

adequate frameworks to the nature of the disease.

Identification, description, and quantification of

the components of a disease cycle are foundational

to plant disease.

Figure 4: Architecture of the predictive analysis

system-Health Care Application

ICCSRE 2018 - International Conference of Computer Science and Renewable Energies

4 CONCLUSION

The health industry has progressed due to the

development of new computer technologies which

gave birth to multiple fields of research. In this

article, we are proposing a system that will help in

reducing the death rate by providing preventive pre-

treatment so that the patient is cured even before

falling ill. The future vision is to implement this

predictive system based on real data

REFERENCES

V. Kirubha and S. Manju Priya, “Survey on data mining

algorithms in disease prediction,” International Journal

of Computer Trends and Technology, vol. 38, no. 3,

pp. 24_128, 2016.

R. Miotto, F.Wang, S.Wang, X. Jiang, and J. T. Dudley,

“Deep learning for healthcare: review, opportunities

and challenges”, Briefings in bioinformatics, 2017.

M. Mohammadi, A. Al-Fuqaha, S. Sorour, and M.

Guizani, “Deep learning for iot big data and streaming

analytics: A survey”, arXiv preprint

arXiv:1712.04301, 2017.

L. Deng, “A tutorial survey of architectures, algorithms,

and applications for deep learning,” APSIPA

Transactions on Signal and Information Processing,

vol. 3, 2014.

Y. Jia, E. Shelhamer, J. Donahue, S. Karayev, J. Long, R.

Girshick, S. Guadarrama, and T. Darrell, “Caffe:

Convolutional architecture for fast feature

embedding,” in Proceedings of the 22nd ACM

international conference on Multimedia. ACM, 2014,

pp. 675_678.

S. D. Arasu and R. Thirumalaiselvi, “Review of chronic

kidney disease based on data mining techniques,”

International Journal of Applied Engineering

Research, vol. 12, no. 23, pp. 13 498_13 505, 2017.

Y. Zhang, M. Qiu, C.-W. Tsai, M. M. Hassan, and A.

Alamri,“Health-cps: Healthcare cyber-physical system

assisted by cloud and big data,” IEEE Systems

Journal, vol. 11, no. 1, pp. 88_95,2017.

S. Bahrampour, N. Ramakrishnan, L. Schott, and M. Shah,

“Comparative study of caffe, neon, theano, and torch

for deep learning,” 2016.

M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean,

M. Devin, S. Ghemawat, G. Irving, M. Isard et al.,

“Tensorflow:A system for large-scale machine

learning.” in OSDI, vol. 16, 2016, pp. 265_283.

D.Ravi, C.Wong, F. Deligianni, M. Berthelot, J. Andreu-

Perez, B. Lo, and G.-Z. Yang, “Deep learning for

health informatics,” IEEE journal of biomedical and

health informatics, vol. 21, no. 1, pp. 4_21, 2017.

R. L. Dumitru, “Iot platforms: Analysis for building

projects,” Informatica Economica,

Predictive Analytics in Healthcare System using Deep Learning Approach