Boosting Kidney Stone Identiﬁcation in Endoscopic Images Using Two-Step Transfer Learning Francisco Lopez-Tiro123 Juan Pablo Betancur-Rengifo3 Arturo Ruiz-Sanchez3 Ivan Reyes-Amezcua34

2025-05-06 0 0 2.66MB 5 页 10玖币

侵权投诉

Boosting Kidney Stone Identiﬁcation in Endoscopic

Images Using Two-Step Transfer Learning

Francisco Lopez-Tiro1,2,3, Juan Pablo Betancur-Rengifo3, Arturo Ruiz-Sanchez3, Ivan Reyes-Amezcua3,4,

Jonathan El-Beze5, Jacques Hubert5, Michel Daudon6, Gilberto Ochoa-Ruiz*,1,3, Christian Daul*,2

Abstract—Knowing the cause of kidney stone formation is

crucial to establish treatments that prevent recurrence. There are

currently different approaches for determining the kidney stone

type. However, the reference ex-vivo identiﬁcation procedure can

take up to several weeks, while an in-vivo visual recognition

requires highly trained specialists. Machine learning models

have been developed to provide urologists with an automated

classiﬁcation of kidney stones during an ureteroscopy; however,

there is a general lack in terms of quality of the training

data and methods. In this work, a two-step transfer learning

approach is used to train the kidney stone classiﬁer. The proposed

approach transfers knowledge learned on a set of images of

kidney stones acquired with a CCD camera (ex-vivo dataset)

to a ﬁnal model that classiﬁes images from endoscopic images

(ex-vivo dataset). The results show that learning features from

different domains with similar information helps to improve

the performance of a model that performs classiﬁcation in real

conditions (for instance, uncontrolled lighting conditions and

blur). Finally, in comparison to models that are trained from

scratch or by initializing ImageNet weights, the obtained results

suggest that the two-step approach extracts features improving

the identiﬁcation of kidney stones in endoscopic images.

Index Terms—Transfer learning, kidney stones, deep learning

I. INTRODUCTION

The formation of kidney stones in the urinary tract is a

public health issue [1], [2]. In industrialized countries, 10%

of the population suffers from an episode of kidney stones

during their lifetime. Recent studies have determined that the

risk of recurrence increases up to 40% in less than 5 years

[3], [4]. Thus, determining the root cause of kidney stone

formation is crucial to avoid relapses through personalized

treatments [3], [5], [6]. Therefore, different approaches for

visually identifying some of the most common types (or

classes) of kidney stones have been proposed in recent years

[7], [8].

The Morpho-Constitutional Analysis (MCA) is currently

the reference method for the identiﬁcation of the type of the

extracted kidney stone fragments [9]. This ex-vivo procedure

consists of two complementary analyses on the extracted

kidney stone parts, which were fragmented with a laser.

1Tecnologico de Monterrey, School of Sciences and Engineering, Mexico

2CRAN (UMR 7039, Universit´

e de Lorraine and CNRS), Nancy, France

3CV-INSIDE Lab Member, Mexico

4CINVESTAV, Computer Sciences Department, Mexico

5CHU Nancy, Service d’urologie de Brabois, Vandœuvre-l`

es-Nancy, France

6Hˆ

opital Tenon, AP-HP, Paris, France

*Corresponding authors:

gilberto.ochoa@tec.mx, christian.daul@univ-lorraine.fr

The fragments are visually inspected under a microscope to

observe the colors and textures of their surface and sec-

tion. Then, an infrared-spectrophotometry analysis enables

to identify the molecular and crystalline composition of the

different areas (layers) of the kidney stone [10]. However,

in numerous hospitals the MCA results are only available

after some weeks. This delay makes it difﬁcult to establish an

immediate and appropriate treatment for the patient. On the

other hand, removing large kidney stone fragments is often

difﬁcult in practice. Moreover, the biochemical composition

can be altered by the laser during the fragmentation [11],

making the MCA procedure challenging in some cases.

Endoscopic Stone Recognition (ESR) is a promising tech-

nique to immediately determine the type of kidney stones dur-

ing the ureteroscopy (i.e., in-vivo recognition). The advantage

of ESR is twice: kidney stones can be pulverized (dusting

procedure with a laser) instead fragmented, and an appropriate

treatment can be immediately deﬁned. ESR is only based on

a visual inspection of in-vivo endoscopic images observed

on a screen. For trained urologists, ESR results are strongly

correlated with those of MCA [12]. However, only a few

highly trained experts are nowadays able to recognize the type

of kidney stones using only endoscopic images. Moreover, the

visual classiﬁcation by urologists is operator dependent and

subjective, and the required experience is long to acquire [13].

Studies have been recently proposed to automate ESR [12],

[14], [15]. These Deep Learning (DL) based methods led

to promising results. However, one of the most common

challenges in these DL-based methods for classifying kidney

stones is the lack of a large image set for the model training.

In addition, the similarity of the data distribution is another

important factor to obtain an adequate model. Consequently,

this suggests a trade-off between the amount of available

data and the data distribution to ﬁt the network weights

adequately. The majority of the DL-based models report ﬁne-

tuning of weights learned from distributions other than those

from kidney stone images (commonly from ImageNet [16]).

Transfer Learning (TL) is used when features learned from

a given domain (or class of images) can bring appropriate

knowledge to another domain for which the available image

set is too small to train a large model from scratch [17], [22].

In the context of ureteroscopy, a large dataset of in-vivo images

is currently not available and collecting such a large database

of endoscopic images during ureteroscopies is a long term

work. However, in the context of this work, images of ex-vivo

arXiv:2210.13654v1 [cs.CV] 24 Oct 2022

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：

BoostingKidneyStoneIdenticationinEndoscopicImagesUsingTwo-StepTransferLearningFranciscoLopez-Tiro1;2;3,JuanPabloBetancur-Rengifo3,ArturoRuiz-Sanchez3,IvanReyes-Amezcua3;4,JonathanEl-Beze5,JacquesHubert5,MichelDaudon6,GilbertoOchoa-Ruiz*;1;3,ChristianDaul*;2AbstractKnowingthecauseofkidneystoneforma...

展开>> 收起<<

Boosting Kidney Stone Identiﬁcation in Endoscopic Images Using Two-Step Transfer Learning Francisco Lopez-Tiro123 Juan Pablo Betancur-Rengifo3 Arturo Ruiz-Sanchez3 Ivan Reyes-Amezcua34.pdf

共5页,预览1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！

Boosting Kidney Stone Identiﬁcation in Endoscopic Images Using Two-Step Transfer Learning Francisco Lopez-Tiro123 Juan Pablo Betancur-Rengifo3 Arturo Ruiz-Sanchez3 Ivan Reyes-Amezcua34

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: