Deep Learning for Diagonal Earlobe Crease Detection Sara L. Almonacid-Uribe a Oliverio J. Santana

2025-05-06 0 0 1.52MB 8 页 10玖币

侵权投诉

Deep Learning for Diagonal Earlobe Crease Detection

Sara L. Almonacid-Uribe a, Oliverio J. Santana b,

Daniel Hern´

andez-Sosa c, and David Freire-Obreg´

on d

SIANI, Universidad de Las Palmas de Gran Canaria, Las Palmas de Gran Canaria, Spain

david.freire@ulpgc.es

Keywords: Computer vision, diagonal earlobe crease, DELC, Frank’s Sign, cardiovascular disease, coronary artery

disease, deep learning.

Abstract: An article published on Medical News Today in June 2022 presented a fundamental question in its title: Can an

earlobe crease predict heart attacks? The author explained that end arteries supply the heart and ears. In other

words, if they lose blood supply, no other arteries can take over, resulting in tissue damage. Consequently,

some earlobes have a diagonal crease, line, or deep fold that resembles a wrinkle. In this paper, we take a

step toward detecting this speciﬁc marker, commonly known as DELC or Frank’s Sign. For this reason, we

have made the ﬁrst DELC dataset available to the public. In addition, we have investigated the performance of

numerous cutting-edge backbones on annotated photos. Experimentally, we demonstrate that it is possible to

solve this challenge by combining pre-trained encoders with a customized classiﬁer to achieve 97.7% accuracy.

Moreover, we have analyzed the backbone trade-off between performance and size, estimating MobileNet as

the most promising encoder.

1 Introduction

According to the Centers for Disease Control and Pre-

vention (CDC), heart disease is the leading cause of

death for men, women, and the majority of racial

and ethnic groups in the United States (CDC, 2022).

Overall, cardiovascular disease is responsible for one

death every 34 seconds in the United States. Further-

more, one in ﬁve heart attacks are silent; the dam-

age is done, but the individual is unaware (Tsao et al.,

2022). Early detection is essential for providing treat-

ment to alleviate symptoms, reduce mortality, and en-

hance the quality of life (Boudoulas et al., 2016).

As a standard practice, clinicians are taught to di-

agnose coronary artery disease (CAD) based on the

medical history, biomarkers, raw scores, and phys-

ical examinations of individual patients, which they

interpret based on their clinical experience. However,

this approach has evolved due to technological ad-

vances. In the past decade, deep learning (DL) has

demonstrated a promising ability to detect abnormal-

ities in computed tomography (CT) images (Ardila

et al., 2019). Several DL techniques have been pro-

ahttps://orcid.org/0000-0001-6660-0867

bhttps://orcid.org/0000-0001-7511-5783

chttps://orcid.org/0000-0003-3022-7698

dhttps://orcid.org/0000-0003-2378-4277

Figure 1: Celebrities exhibiting a DELC marker. In 1987,

the former CNN interviewer Larry King suffered a heart at-

tack and underwent bypass surgery (photo by Eva Rinaldi,

Wikimedia Commons, CC-BY-SA 2.0). In 2009, the for-

mer comedian and actor Robin Williams underwent aortic

valve replacement surgery (photo by Angela George, Wiki-

media Commons, CC-BY 3.0). The ear is highlighted in

both pictures.

posed to automatically estimate CAD markers from

CT images. The majority of these models predict clin-

ically relevant image features from cardiac CT, such

as coronary artery calciﬁcation scoring (Isgum et al.,

2012; Wolterink et al., 2015; Zeleznik et al., 2021),

non-calciﬁed atherosclerotic plaque localization (Ya-

arXiv:2210.11582v4 [cs.CV] 7 Feb 2023

mak et al., 2014; Zhao et al., 2019), and stenosis from

cardiac CT (Lee et al., 2019; Zreik et al., 2019).

Even though the development of DL on CT im-

ages is promising, CT equipment is expensive and

cardiac illnesses are hard to ﬁnd unless the patient

has symptoms and goes to the hospital for a cardiac

checkup. In this context, the diagonal earlobe crease

(DELC) can be a helpful guide to identify cardiac

problems. This crease extends diagonally from the

tragus to the earlobe’s border (see Figure 1). It is

also known as Frank’s Sign because it was ﬁrst de-

scribed by Frank in a case series of CAD patients

(Frank, 1973). Since then, numerous reports have

been published concerning its association primarily

with atherosclerosis, particularly CAD (Wieckowski,

2021). While not as well known as more traditional

approaches, DELC examinations are painless, non-

invasive, and simple to interpret. If its diagnostic ac-

curacy is sufﬁcient for decision-making, it could be

utilized in primary care or emergency departments.

In this work, we have created a DELC detector

using state-of-the-art (SOTA) backbones and ear col-

lections as benchmarks for the models. First, we have

gathered DELC ear images available on the Internet.

Then, we developed multiple DL models consider-

ing pre-trained encoders, also known as backbones,

to predict whether or not an ear displays a DELC. In

addition, we analyzed the performance of the consid-

ered backbones by varying the classiﬁer parameters

and found no correlation between the number of pa-

rameters and the best model.

Our proposal was evaluated using a mixed-source

dataset. As previously stated, we gathered 342 pos-

itive DELC images by collecting publicly available

images from the Internet and cropping off the ears.

Negative samples were obtained from a publicly ac-

cessible ear database, namely AWE (Emerˇ

siˇ

c et al.,

2017). All images are collected from natural settings,

including lighting, pose, and shape variations. Con-

sidering the number of samples, data augmentation

techniques were considered during training. The out-

comes are remarkable (predictions up to 97.7% accu-

rate) and have yielded intriguing insights.

Even though the earlobe is a relatively small

part of the ear (see Figure 2), the classiﬁer’s perfor-

mance is noteworthy. Unlike other diseases, such as

melanoma, which can be found anywhere in the hu-

man body, the DELC is located in a speciﬁc area,

facilitating the detection task. The usability of the

trained models reveals an additional insightful rev-

elation. Light-weight convolutional neural networks

such as MobileNet provide high accuracy, balancing

the precision of complex neural network structures

with the performance constraints of mobile runtimes.

Helix

Tragus

Lobule

Anti-Helix

Concha

Figure 2: Outer ear scheme. The human earlobe (lobulus

auriculae) is composed of areolar and adipose connective

tissues that lack rigidity. Due to the absence of cartilage in

the earlobe, it has an abundant blood ﬂow and may aid in

warming the ears (Steinberg and Rosner, 2003).

Hence, ubiquitous applications could take advantage

of this proposal, making it possible to detect DELC by

just using a smartphone anywhere and anytime. Our

contributions can be summarized as follows:

• We propose a novel dataset with DELC ear im-

ages in the wild with 342 samples. All samples

have been gathered from the Internet. The dataset

is publicly available.

• We experimentally demonstrate that it is possible

to tackle this problem by combining pre-trained

backbones with a new classiﬁer.

• In this experiment, eleven different backbones are

compared to one another regarding their DELC

detection performance. Moreover, the models’

size-performance trade-off analysis demonstrates

that the problem can be effectively addressed by

employing light-weight encoders. As aforemen-

tioned, this opens the door for the broad imple-

mentation of this technology.

The remainder of this paper is organized as fol-

lows. Section 2 discusses previous related work. Sec-

tion 3 describes the proposed pipeline. Section 4 re-

ports the experimental setup and the experimental re-

sults. Finally, conclusions are drawn in Section 5.

2 Related work

The state of the art can be studied from both phys-

iological and technological viewpoints. The former

aims to ﬁnd support for the relationship between CAD

and DELC by examining related studies, while the lat-

ter intends to evaluate the Computer Vision Commu-

nity proposals.

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：

DeepLearningforDiagonalEarlobeCreaseDetectionSaraL.Almonacid-Uribea,OliverioJ.Santanab,DanielHern´andez-Sosac,andDavidFreire-Obreg´ondSIANI,UniversidaddeLasPalmasdeGranCanaria,LasPalmasdeGranCanaria,Spaindavid.freire@ulpgc.esKeywords:Computervision,diagonalearlobecrease,DELC,Frank'sSign,cardiovascul...

展开>> 收起<<

Deep Learning for Diagonal Earlobe Crease Detection Sara L. Almonacid-Uribe a Oliverio J. Santana.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！

Deep Learning for Diagonal Earlobe Crease Detection Sara L. Almonacid-Uribe a Oliverio J. Santana

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: