A Transfer Learning Based Approach for Classiﬁcation of COVID-19 and Pneumonia in CT Scan Imaging

2025-04-30 0 0 1022.62KB 8 页 10玖币

侵权投诉

A Transfer Learning Based Approach for

Classiﬁcation of COVID-19 and Pneumonia in CT

Scan Imaging

Gargi Desaiγ, Nelly Elsayed†, Zag Elsayed‡, Murat Ozer∗

School of Information Technology

University of Cincinnati

Cincinnati, Ohio, United States

γdesaigd@mail.uc.edu, †elsayeny@ucmail.uc.edu, ‡elsayezs@ucmail.uc.edu, ∗ozermm@ucmail.uc.edu

Abstract—The world is still overwhelmed by the spread of

the COVID-19 virus. With over 250 Million infected cases as

of November 2021 and affecting 219 countries and territories,

the world remains in the pandemic period. Detecting COVID-

19 using the deep learning method on CT scan images can

play a vital role in assisting medical professionals and decision

authorities in controlling the spread of the disease and providing

essential support for patients. The convolution neural network

is widely used in the ﬁeld of large-scale image recognition.

The current method of RT-PCR to diagnose COVID-19 is time-

consuming and universally limited. This research aims to propose

a deep learning-based approach to classify COVID-19 pneumonia

patients, bacterial pneumonia, viral pneumonia, and healthy

(normal cases). This paper used deep transfer learning to classify

the data via Inception-ResNet-V2 neural network architecture.

The proposed model has been intentionally simpliﬁed to reduce

the implementation cost so that it can be easily implemented

and used in different geographical areas, especially rural and

developing regions.

Index Terms—Transfer learning, image classiﬁcation, CT scan,

deep learning

I. INTRODUCTION

Emerging and resurfacing bacteria are universal challenges

for human health. Coronavirus is a contagious respiratory ill-

ness caused by severe acute respiratory syndrome coronavirus

2 (SARS-CoV-2), which is responsible for the COVID-19

pandemic. The ﬁrst case of novel coronavirus was reported

in Wuhan, Hubei province, China, in December 2019 [1].

The examination of the disease suggests that the outbreak was

associated with the seafood market in Wuhan. Coronavirus is

a closed Ribonucleic Ac-id (RNA) that is classiﬁed among hu-

mans, other mammals, and birds. There are six types of coro-

navirus species that infect humans. Four of these viruses that

commonly cause the common cold are 229E, OC43, NL63,

and HKU1 [2]. The other two viruses, namely SARS-COV

and MERS-COV, are linked to humans’ fatal illnesses [3].

The World Health Organization (WHO) declared the outbreak

of COVID-19 pneumonia as a pan-demic in March 2020.

Globally, as of 10th June 2021, there have been 174,061,995

conﬁrmed cases of COVID-19, including 3,758,560 deaths,

reported to the World Health Organization (WHO) [4]. The

symptoms of the disease are varied. Common symptoms

include respiratory illness, cough, headache, loss of smell

and taste, nasal congestion, muscle pain, sore throat, fever,

and diarrhea. The presence of the virus in the human body

was identiﬁed by using sequencing in samples from patients

with pneumonia [2]. The present diagnosis for coronavirus

includes reverse-transcription polymerase chain reaction (RT-

PCR), real-time RT-PCR (rRT-PCR), and reverse transcription

loop-mediated isothermal ampliﬁcation (RT-LAMP). Other

tests also involve nasopharyngeal and oropharyngeal swab

tests to detect COVID-19. However, these tests are time-

consuming, and the shortage of kits delays the diagnosis [5].

Other than these tests, computed tomography (CT) scans and

chest X-rays (CXR) are used to detect COVID-19 pneumonia.

The COVID-19 pandemic is still a primary challenge for the

healthcare sector, especially in rural and developing regions

with a signiﬁcant shortage of medical personnel.

Artiﬁcial intelligence (AI) attempts to build intelligent mod-

els and includes machine learning and deep learning sub-

sets. Deep learning methods can enhance the image features

which are not visible in the original image [6]. Computed

tomography of the chest utilizes X-ray apparatus to investigate

abnormalities found in tomography tests and to help diagnose

the reason for the shortness of breath, fever, chest pain, and

other chest symptoms. Computed tomography is accurate,

painless, and non-invasive, and it can detect tiny nodules in

the lung. Early detection of cases of COVID-19 pneumonia

for timely treatment is crucial for avoiding the spread of the

epidemic [7]. Nonetheless, this remains a difﬁcult task to be

completed. A vast number of patients across different locations

and the limited medical resources cause a delay in early

detection, resulting in delayed decisions on hospitalizations,

which increases the chances of cross-infections and poor

prognosis. The current RT-PCR test to determine COVID-19

infection has some limitations. The current test (RT-PCR) is

not available universally, the processing time can be lengthy,

and the sensitivity report varies. As new studies come up, we

cannot solely rely on the RT-PCR test for diagnosing COVID-

19, especially in patients without symptoms [8], [9]. It is

challenging to distinguish between the CT scans of COVID-19

pneumonia patients and general pneumonia patients because

arXiv:2210.09403v2 [eess.IV] 26 Oct 2022

Fig. 1. Inductive learning versus inductive transfer methodology.

Fig. 2. CT scan images showing ground-glass opacities in con-ﬁrmed COVID-

19 positive patient.

both viruses belong to the same Coronaviridae family. In a

cohort of 1014 patients, the ratio of positive chest CT scan

results is 88%, and that of RT-PCR test is 59% [5]. Although

RT-PCR is most often in agreement, CT can detect COVID-

19 in negative RT-PCR tests in people with no symptoms,

and it has been an integral part of diagnosis in multiple

centers in Wuhan, China, and northern Italy [10]. A recent

international expert’s report bolsters the use of chest CT for

COVID-19 patients with deteriorating respiratory conditions or

in a restricted resource environment for medical classiﬁcation

of patients who show average to severe clinical traits [8].

Given the trend for true artiﬁcial general intelligence,

transfer learning is something researchers believe can further

our progress towards strong AI [11]. While there might be

advanced models, with high precision and knocking all stan-

dards, they would be only on very particular datasets and end

up with a loss in performance when used in a new task which

might still be like the one it was trained on [11]. This shapes

the drive for transfer learning, which goes beyond functions

and domains, and tries to leverage knowledge from pre-trained

models and use it to solve new problems. In the paper [12],

the authors used domain, task, and marginal probabilities to

present a framework for understanding transfer learning [12].

The framework is deﬁned as follows: A domain, D, is de-

ﬁned as a two-element tuple consisting of feature space, ϕ,

and marginal probability, P(ϕ), where ϕis a sample data

point [12]. Thus, we can represent the domain mathematically

as D = ϕ, P(ϕ) [13]. Inductive transfer techniques utilize the

inductive biases of the initial task to assist the goal task.

This can be done in different ways, such as by adjusting the

inductive bias of the target task by limiting the model space,

narrowing down the hypothesis space, or adjusting the search

process itself with the help of knowledge from the original

task [11]. This process is depicted visually in Fig. 1.

The main intention of this study is to provide an assist-

ing model for medical personnel to classify CT scans. The

proposed research aims to differentiate COVID-19 pneumonia

from bacterial pneumonia, viral pneumonia, and normal lungs

with no disease from chest CT scans. The model intentionally

targets the model design to be simple and accessible to the

rural and developing regions where there is an overall shortage

of medical personnel, especially lung disease professionals.

Moreover, it can be used in any region to support professional

medical personnel as a second opinion that can support or

discover unseen features in the CT scan so that the medical

decision can be revised or request additional tests for the

patient to perform the most precise diagnoses. In this study, we

use the Inception-Res-Net-V2 pre-trained model for transfer

learning. In addition, we used publicly available COVID-19

and the pneumonia dataset [7] for easy replication of the

model.

II. RELATED WORK

Deep learning is a subset of machine learning involved

with algorithms stimulated by the shape and role of the brain

called artiﬁcial neural networks (ANNs). In simple words,

deep learning can be thought of as a way to computerize

advanced analytics that uses both new and historical data to

predict activity. Similar to how humans learn from knowledge

and experience gained over time, the deep learning model

would perform a task many times, each time adjusting it a little

to enhance the output. Deep learning is a signiﬁcant compo-

nent of data science, which includes advanced analytics [14],

[15] and predictive modeling [16]–[18]. Various techniques

are used to create strong deep learning models, including

learning rate decay, training from scratch, transfer learning,

and dropout. To date, deep learning is rising as the foremost

machine-learning tool in the general imaging and computer

vision domains [13]. CNNs have been demonstrated and tested

to be powerful tools for a vast extent of computer vision

tasks. Studies indicate that medical image analysis groups are

quickly adapting to the ﬁeld of CNNs and other deep learning

techniques to a wide variety of applications [13]. In medical

diagnostics, the assessment of disease relies on both image in-

terpretation and image acquisition. With evolving technology,

image attainment has improved substantially over recent years,

with improved devices to capture increased-resolution images.

However, image interpretation has recently begun to beneﬁt

from computer vision technology. Most of the medical image

interpretations are performed by physicians. These human-

intervened interpretations are limited to subjectivity, with large

variations across physicians. Deep learning has proven to be

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：

ATransferLearningBasedApproachforClassicationofCOVID-19andPneumoniainCTScanImagingGargiDesai,NellyElsayedy,ZagElsayedz,MuratOzerSchoolofInformationTechnologyUniversityofCincinnatiCincinnati,Ohio,UnitedStatesdesaigd@mail.uc.edu,yelsayeny@ucmail.uc.edu,zelsayezs@ucmail.uc.edu,ozermm@ucmail.uc.eduAb...

展开>> 收起<<

A Transfer Learning Based Approach for Classiﬁcation of COVID-19 and Pneumonia in CT Scan Imaging.pdf

共8页,预览2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！

A Transfer Learning Based Approach for Classiﬁcation of COVID-19 and Pneumonia in CT Scan Imaging

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: