Investigation of Early -Stage Breast Cancer Detection using Quantum Neural Network Musaddiq Al Ali a Amjad Y. Sahib b Muazez Al Ali c a Department of Advanced Science and Technology Toyota Technological Institute 2 -12-1Hisakata Tenpaku -ku

2025-05-03 0 0 862.22KB 17 页 10玖币

侵权投诉

Investigation of Early-Stage Breast Cancer Detection using Quantum Neural Network

Musaddiq Al Ali a, Amjad Y. Sahib b, Muazez Al Ali c

a Department of Advanced Science and Technology, Toyota Technological Institute, 2-12-1,Hisakata, Tenpaku-ku,

Nagoya, Aichi 468-8511, Japan

b University of Wasit, College of engineering. Al Kut, Wasit, Iraq

cAl Ayen University, College of Dentistry, Nile St, Nasiriyah, Iraq

Corresponding author: alali@toyota-ti.ac.jp (Musaddiq Al Ali)

Abstract

Computer-aided image diagnostics (CAD) have been used in many fields of diagnostic medicine. It relies heavily on

classical computer vision and artificial intelligence. Quantum neural network (QNN) has been introduced by many

researchers around the world and presented recently by research corporations such as Microsoft, Google, and IBM. In

this paper, the investigation of the validity of using the QNN algorithm for machine-based breast cancer detection was

performed. To validate the learnability of the QNN, a series of learnability tests were performed alongside with classical

convolutional neural network (CCNN). QNN is built using the Cirq library to perform the assimilation of quantum

computation on classical computers. Series of investigations were performed to study the learnability characteristics of

QNN and CCNN under the same computational conditions. The comparison was performed for real Mammogram data

sets. The investigations showed success in terms of recognizing the data and training. Our work shows better

performance of QNN in terms of successfully training and producing a valid model for smaller data set compared to

CCNN.

Keywords: Quantum neural network, Breast cancer, Classical neural network, Machine learning, Mammography.

1- Introduction

With the advancement in medical and engineering fields, novel solutions were implemented to facilitate patients'

health care and prolong their life[1,2,11–20,3,21–26,4–10] . The use of computer use of computer-aided diagnostic

(CAD) is an important topic in engineering-medical research[27,28]. Recently, many researchers investigated the

concept of automating CAD by building self-learning algorithms based on machine learning [29][30][31]. In order to

build a successful diagnostic model of medical images, classical machine learning is used. However, training classical

machine learning is consuming a huge computational resource in terms of the data set preparation as well as computer

resources for the training phase [31][32]. For the medical diagnostic field, data are mostly visual-based, such as X-rays,

computed tomography scan and magnetic resonance imaging (MRI), etc. Therefore; computer vision tools are the most

appropriate method to be used for CAD. Add to that, most data in modern diagnostic are computerized based [33]. As

such, artificial intelligence is the most suitable to be used in the next generation of fully computerized CAD.

Historically, CAD [34] has been approached in two sequential steps. The first step is to screen the data to detect the

suspicious regions or what are so-called (Regions of interest) then, the region of interest will be “labeled” for the closest

possible medical cases according to the corresponding disease likelihood. This is done by what so called expert systems

[35][36]. Expert systems will assign diagnostic labels according to the probability of the region of interest diagnostic

ascendingly, then eliminate medical cases of lower probabilities and then identify the highly likely disease. The problem

with the use of expert systems is that they are built upon predetermined diagnostics for a fixed small number of cases

and are mostly based on a wide variety of collective diagnostic data. For example, to diagnose a single case of apparent

mass in appeared in mammography, a series of tests should be performed (Blood, tissue samples, multiple scans, etc.)

to be fed to the expert systems than gives the preliminary diagnosis. On the other hand, artificial intelligence (AI) has

great potential to replace traditional expert systems and allows one to reach a preliminary diagnosis in a very short time.

AI is presented itself as a powerful tool for image classification and medical identification due to its characteristics of

transforming representing information sets of data (database) into structured matrices of simple units containing

weighted partial differential equations (PDEs). Weights then will be tuned to draw learning pathways as an intuitive

mimic of the learning process of the biological neural system (i.e., the brain). The satisfying amount of learning data

(i.e., raining data) to provide a robust outcome or valid model is depending on the AI model design and the problem

itself. For data selection, two major aspects should be taken into consideration; the first one is identifying the objective

of data to be trained for. In other words, what are the systematic methodologies of the doctor to make diagnostics based

on such data. For example, in microscopic cultures, counting cells is one of the diagnostic tools; therefore, counting will

necessitate isolating (through image segmentation) as well as labeling the targeted cells in each picture before submitting

the data to the deep learning model. The second aspect is the data set size and number of items for each label. As it has

been expressed before, training robustness is susceptible to data set size and the balance of data distribution in each

label. The bottleneck of creating a strong and highly trusted CAD is the computational power. For classical machine

learning (which is performed on nowadays classical computers that use the binary system [0,1]), the researchers reach

the computation limitations to modeling real-life problems such as biochemical interactions and immune system

interaction with infection. Some algorithms are invented to rework the data feed and training process to optimize

workflow for the available resources. Furthermore, some hardware-based methods are introduced to give the classical

systems the needed dynamic storage to allocate the data feeds for the processing units as well as storing the tensor from

the data (such as Intel Obtain-based modules and tensor processors). However, the largest computer (supercomputer) is

still facing huge challenges to simulate the simulation tasks mentioned earlier. As such, in the medical field, classical

computers are still facing serious limitations in terms of attaining good diagnoses compared to well-trained doctors with

the same amount of data. To overcome classical computers’ limitations, researchers are working on developing so-

called quantum computers. Quantum computers are applying the principles of entanglement and superpositions to the

data unit, i.e., the bit to build the quantum bit. There are several approaches to physically implementing quantum

computers such as photonic and silicon-based circuitry. However, the development is not moving fast as it has been

anticipated due to entropy and noise problems. On the other hand, the logical aspects of a quantum computer are mature

enough to be implemented as soon as a full-scale quantum computer is available. The advantages of a quantum computer

can be shown by the phenomenal calculation speed and the data amount to be handled in a single calculation. The high

speed and huge data processing ability of quantum computers are due to the nature of the method of conveying

information itself, such that, instead of the binary representation of the data (i.e., classical bit), quantum bits within the

quantum gates exchange information timelessly. The property of a quantum state makes one quantum computer hold

calculation power equivalent to all existing classical computers (i.e., quantum supremacy) [37].

Quantum computers and computation are superior due to the nature of quantum information and quantum logic.

Although the quantum computation field of study is relatively a new branch of applied mathematics, however; the

physical and mathematical advancements of quantum computers are motivating the researchers to race toward

implementing a unique type of logic gates and physical hardware. Nowadays, limited quantum bits computers are

already serving in specific high technology applications such as pharmacy, and chemical interactions in next-generation

batteries. It is anticipated that quantum computers will be available for public use with full capability at the end of the

21st century. The use of quantum computers for medical diagnostics can play a vital role in terms of the implementation

of fast and systematic machine-based diagnostics of malignant tumors which are treatable if they can be diagnosed in

the early stages. Breast cancer is the most frequent malignant tumor amongst women. It is a dominant cause of female

mortality and is considered a serious public health problem all over the world. Current treatments for breast cancer

include surgery, chemotherapy,

immunotherapy, and radiation therapy. Breast cancer incidence and death rates increase with age but are mortality rate

decrease significantly if it can be detected in the early stages, before the metastasizing phase. The eradication and

therapeutic success of breast cancer are related to tumor stratification and dissemination. Breast tumors whether they

are benign or malignant are distinguished into four major classes, based on size, age, node involvement, and tumor

grade. These stages are 1; consists of the well-defined and localized tumor mass, characterized by poor invasion

properties. Stage 2 and 3, corresponds to an increased tumor volume and acquisition of invasive phenotype. The

metastasis dissemination and huge tumor size with invasive phenotype are classified as stage 4. Chemotherapy,

radiation, and targeted therapies have made major advances in patient management over the past decades, but refractory

diseases and recurrence remain common. The early-stage diagnostic will lead to treating cancer before the metastasis

stage, at which cancer will attack different organs by migrating through lymph nodes. A mammogram is one of the

popular tests for breast cancer early detection. Mammograms have been used efficiently to reduce the mortality rate of

women with breast cancer. Early detection is based on the oncologist's exam of the x-ray image and then examining the

suspicious tissue by taking a biopsy. However, due to the limitation of highly trained medical staff, cancer false positive

is common in mammogram image detection as well as false negative [38–40] due to the misconception of less-skilled

eyes for the masses in the image. Another point worth mentioning is that mammogram is an X-ray image that take short

time to be made, yet the speed of mass examination is related to how many doctors exist and their level of experience.

Therefore, CAD is a promotion point as a necessary way to reduce diagnostic time [41–43] and decrease cost per test

by reducing the number of specialists in the mammogram testing unit in hospitals (This reduction will lead to

remobilizing the manpower to other sections within the hospital which will lead to double the efficiency of the hospital’s

workforce) as well as the reduction of the number of false-positive by double-checking the X-ray image by the doctor

and the computer.

This paper is examining the use of QNN to build an “ultra-intelligent machine”. To attain this goal, the following

Key questions should be answered: Firstly, how to build a QNN and how to transform the classical data from plane

mammogram images to become quantum data. The second key question is how to evaluate the learnability of QNN.

The third question is, how is implementing a benchmarking of learnability evaluation of QNN? The final question is:

can we deduce that, QNN is able to perform successful mammography diagnostic in the future according to the current

investigation?

As such, this study introduced the learnability factors. Furthermore, a series of numerical investigations were

conducted to examine the various aspects that govern AI with QNN. Moreover, a hybrid model of CCNN and QNN is

introduced to allow the implementation of quantum logic in the near future instead of waiting for a fully capable

quantum computer to conduct breast cancer early detection for mammography.

the layout of the paper is the following. Section 2 is discussing breast cancer and diagnostic principles. whereas

Section 3 deals with the learning of classical and quantum neural networks. Section 4 focuses on quantum neural

network learning. Section 5 discusses the layout of quantum and classical neural networks and mammogram data

structure. In section 6, we will present the results and discuss the outcome. Finally, the conclusions are presented in

section 7.

2- Brief of breast cancer biology and transcriptional regulation

Breasts are made up of connective, glandular, and fatty tissues that have lobes, lobules, ducts, areola, and

nipple[44,45]. These organs consist of a uniform structure of epithelial cells that secrete and produce milk after

childbirth. Whenever there is a morphologic or functional alteration within its uniform epithelial structures, tumor

initiation develops and later forms a mass of multiple populations of cells capable of evading physiological cell death.

The changes in gene expression patterns seen in breast cancer[46]have provided evidence of epigenetic, genetic, or post-

translational altered expression of certain proteins, like transcription factors, co-regulators, and histone enzymes that

order DNA into structural units according to a recent study. These proteins play a crucial role in the expression of genes

that results in the susceptibility of a healthy cell to the transformation of a malignant cell. Among the first altered

transcriptional regulation found in breast cancer were the overexpression and gene amplification of estrogen receptor

alpha (ERα) and avian myelocytomatosis viral oncogene homologue factor (c-myc). These two oncoproteins were found

to be associated with abnormal cell division and replication within the breast. Additional studies have identified

inherited/acquired altered gene expression as a detectable cause of carcinogenesis of breast tissue. This arises after a

study of some essential genes involved in cellular processes and maintenance were found to be mutated at the germ cell

level. Next-generation sequencing analysis also found higher penetrance mutations in breast cancer 1 (BRCA1), tumor

protein p53, mitogen-activated protein kinase 1 (MAP3K1), retinoblastoma 1 (RB1), phosphatidylinositol-4, 5-

bisphosphate 3-kinase catalytic subunit alpha (PIK3CA), and GATA binding protein 3 (GATA-3) genes that result in

breast cancer formation. Breast cancer diagnostics has two-stage. The first stage is identifying whether there is

irregularity within breast tissues (Masses). The second stage is doing a biopsy of the suspicious mass. Abnormal masses

within breast tissue are the significant features that determine the likelihood of cancer's existence in the breast. The first

identification is shaped. The shape can be rounded or irregular. An irregular shape increases the likelihood of

malignancy (Cancer). Margins of the masses are also important. The margin can be circumscribed, microlobulated,

obscured (partially hidden by adjacent tissue), indistinct (ill-defined), or spiculated. The likelihood of malignancy of

circumscribed margins is low. The density of the mass is also a factor in early diagnosis. For example, if the mass is

cystic or fibroadenomas mass. Other malignancy signs are neodensity, architectural distortion, or asymmetric density.

3- Classical and quantum neural network

Machine learning is the transcending form of the neural network, which was immersed first in the form of a “logic

theorist program” that was invented in 1956 [47]. The neural network process starts by discretizing the problem to what

so called neurons (Χ in Fig. 1). Multiple neurons (thousands and even millions) then will be stacked in a certain design

to make the machine learning architecture. The information to be processed will be distributed on the neurons within

the machine learning architecture, then fitting the response of each neuron to arbitrary function ψ. Functions output will

be submitted to a comparator Ω which judges the data based on its probability function the gives the generalized output

of the subsystems Ψ.

Fig. 1. Machine intelligence process unit (neuron).

In summary, machine learning is a network of node clusters; each node has a weighted subfunction to be tuned. This

tuneable subfunction is representative of a fraction of the resulting model. The node's output mimics neural cells by

adopting an activation function to control the output with predefined criteria. Activation function can take the form of

rectified linear, Sigmoid, or hyperbolic. The stacks are consisting of an aggregation of neurons in layers. The layers

may take many formations according to the machine learning architecture (e.g., dense and convolution layers). Training

of the neural network is performed by tuning each neuron's weights for the optimum value that fits the training data.

Here, optimization algorithm selection plays a significant role in training success. With increasing the training rate, data

loss for each prediction will drop. If the data is insufficient or the neural network is designed poorly, the fit convergence

cannot reach sufficient value. In this case, the underfitting problem is occurring. Contrary to the underfitting problem,

the overfitting problem occurs when the convergence is satisfied for the model; however, the neural network has the

poor capability to predict and recognize unique and new input data that is not in the training set.

Fig. 2. Under and overfitting of neural network.

Quantum machine learning is a subject of the quantum computation branch, that has been gaining attention in recent

decades, which emerged from quantum computations. Quantum computation has emerged from the statistical

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

InvestigationofEarly-StageBreastCancerDetectionusingQuantumNeuralNetworkMusaddiqAlAlia,AmjadY.Sahibb,MuazezAlAlicaDepartmentofAdvancedScienceandTechnology,ToyotaTechnologicalInstitute,2-12-1,Hisakata,Tenpaku-ku,Nagoya,Aichi468-8511,JapanbUniversityofWasit,Collegeofengineering.AlKut,Wasit,IraqcAlAyen...

展开>> 收起<<

Investigation of Early -Stage Breast Cancer Detection using Quantum Neural Network Musaddiq Al Ali a Amjad Y. Sahib b Muazez Al Ali c a Department of Advanced Science and Technology Toyota Technological Institute 2 -12-1Hisakata Tenpaku -ku.pdf

共17页,预览4页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Investigation of Early -Stage Breast Cancer Detection using Quantum Neural Network Musaddiq Al Ali a Amjad Y. Sahib b Muazez Al Ali c a Department of Advanced Science and Technology Toyota Technological Institute 2 -12-1Hisakata Tenpaku -ku

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: