2022年医疗数据挖掘论文题材.zip资源-CSDN文库

共5个文件

pdf：5个

ieee论文

需积分: 13 80 浏览量 2022-11-08 14:17:26 上传评论收藏 2.23MB ZIP 举报

资源推荐

资源详情

资源评论

收起资源包目录

2022年医疗数据挖掘论文题材.zip （5个子文件）

医疗数据挖掘

Data_Mining_Techniques_for_Medical_Data_A_Review.pdf 155KB

Itani_et_al._(2019).pdf 923KB

1905.09203.pdf 551KB

10.1.1.443.4049.pdf 185KB

procamiaafs00001-0140.pdf 779KB

Speciﬁcs of Medical Data Mining for Diagnosis Aid:

A Survey

Sarah Itani

a,b,∗

, Fabian Lecron

, Philippe Fortemps

Fund for Scientiﬁc Research - FNRS (F.R.S.-FNRS), Brussels, Belgium

Faculty of Engineering, University of Mons, Department of Mathematics and Operations Research, Mons, Belgium

Faculty of Engineering, University of Mons, Department of Engineering Innovation Management, Mons, Belgium

Abstract

Data mining continues to play an important role in medicine; speciﬁcally, for the development

of diagnosis aid models used in expert and intelligent systems. Although we can ﬁnd abundant

research on this topic, clinicians remain reluctant to use decision support tools. Social pressure

explains partly this lukewarm position, but concerns about reliability and credibility are also put

forward. To address this reticence, we emphasize the importance of the collaboration between both

data miners and clinicians. This survey lays the foundation for such an interaction, by focusing on

the speciﬁcs of diagnosis aid, and the related data modeling goals. On this regard, we propose an

overview on the requirements expected by the clinicians, who are both the experts and the ﬁnal

users. Indeed, we believe that the interaction with clinicians should take place from the very ﬁrst

steps of the process and throughout the development of the predictive models, thus not only at the

ﬁnal validation stage. Actually, against a current research approach quite blindly driven by data,

we advocate the need for a new expert-aware approach. This survey paper provides guidelines to

contribute to the design of daily helpful diagnosis aid systems.

Keywords: Data Mining; Medicine; Diagnosis Aid; Explainable Artiﬁcial Intelligence

1. Introduction

As one of the trendiest research topics of our century, Data Mining (DM) makes key contribu-

tions to the scientiﬁc and technological advance in a considerable number of ﬁelds (Gupta, 2014;

PhridviRaj and GuruRao, 2014). Coined during the nineties, the discipline is subject to a tough

competition for the development of algorithms always more powerful, which aim at processing data

∗

Corresponding author. University of Mons, Department of Mathematics and Operations Research, Rue de

Houdain, 9, 7000 Mons, Belgium.

Email addresses: [email protected] (Sarah Itani), [email protected] (Fabian Lecron),

[email protected] (Philippe Fortemps)

200

400

600

800

1000

1200

1990 1995 2000 2005 2010 2015

Number of publications

Year

Figure 1: Evolution of the annual number of publications related to medical data mining in the Scopus database (Sco-

pus) on a quarter of a century, from 1990 to 2015

to infer some knowledge in the form of patterns and/or relationships (Bellazzi and Zupan, 2008).

The associated techniques are derived from the ﬁelds of both statistics and Machine Learning (ML),

the latter which aims at developing computational methods able to extract generalizations from a

set of data (Giudici, 2005).

Medical applications feature among the concerns of the DM community, with a signiﬁcant in-

crease in research interest over the last years (see Figure 1). This interaction comes in diﬀerent

disciplines (Bellazzi et al., 2011): at the cellular and molecular level (bioinformatics); at the tis-

sue and organ level (imaging informatics); at the single patient level (clinical informatics); at the

population and society level (public health informatics).

For half a century now, diagnosis prediction has been a very active research area of clinical

informatics (Wagholikar et al., 2012). In this regard, with the advent of DM, research has progres-

sively shifted away from the statistical approach long considered as a standard practice. Actually,

under a hypothetico-deductive process, statistical analyses are driven to check a hypothesis stated

beforehand and data samples are collected for this special purpose (Yoo et al., 2012). This statis-

tical approach is surely adapted to raise diﬀerences between pathological and control groups, but

not to set an individual assessment, i.e. a clinical examination per subject. In contrast, enriched by

ML techniques, DM inductively processes a voluminous amount of data, to both extract knowledge

and develop predictive models able to help in diagnosing pathologies (Vieira et al., 2017; Yoo et al.,

2012; Bellazzi and Zupan, 2008). In such a process, statistics may ﬁnd its place in feature engineer-

ing, before the stage of model building which is mainly based on ML methods of classiﬁcation or

regression (Esfandiari et al., 2014).

In that respect, it is through data mining that recent works were devoted to the early detection

of cancer, e.g. see Lyu and Haque (2018); Aliˇckovi´c and Subasi (2017); Cichosz et al. (2016); Nahar

et al. (2016); Esfandiari et al. (2014); Krishnaiah et al. (2013);Parvin et al. (2013); Gupta et al.

(2011). Other pathologies, such as cardiac and pulmonary diseases, diabetes, hypertension, meningi-

tis form besides a signiﬁcant part of the research for more precise diagnoses (Esfandiari et al., 2014).

Several psychiatric disorders, such as Attention Deﬁcit Hyperactivity Disorder (ADHD) (Itani et al.,

2018a; Abraham et al., 2017; Milham et al., 2012), Alzheimer (Papakostas et al., 2015), autism (Kos-

micki et al., 2015), schizophrenia, depression and Parkinson (Woo et al., 2017) are also the object

of extensive investigation.

As probably perceived by most of researchers, and certainly by the authors of the present paper,

diagnostic decision support systems that have been proposed so far are not unanimously approved

by clinicians (Wagholikar et al., 2012). Such systems, and the underlying predictive models, are

notably found as being far from the ﬁeld reality. It is thus most likely that data miners are not

enough attentive to the speciﬁcs of medical diagnostic decision support. In particular, though the

DM community was sensitized about the distinctive nature of medical applications (Cios and Moore,

2002), the predictive performance remains practically the lonely parameter within the scope of data

miners, which encourages competition. This trend has been accentuated with the greater availability

of open medical databases, shared by diﬀerent medical and research centers worldwide (Di Martino

et al., 2017; Woo et al., 2017; Di Martino et al., 2014; Esfandiari et al., 2014; Mennes et al., 2013;

Ihle et al., 2012; Kerr et al., 2012; Milham et al., 2012; Poline et al., 2012). Some of these datasets

were launched at the occasion of oﬃcial contests, e.g. the ADHD-200 collection (Milham et al.,

2012). In focusing almost exclusively on performance, these research works (1) miss challenges of

better perceiving and understanding the issues proper to the medical ﬁeld, (2) are exposed to the

risk of yielding inconsistent models, since notably, recent studies showed that there may be no logic

behind the predictions of accurate models (Ribeiro et al., 2016).

It is our strong conviction that the clinicians have to be involved in the whole development

process of their decision support systems. Indeed, they bring expertise and knowledge to contribute

to intelligent and expert systems. That is why, in the present paper, we will shed light upon the

speciﬁcs of medical data mining for diagnosis aid and raise the related data modeling goals. For

such a purpose, we will address the following questions.

(1) How can decision support models be more attractive to clinicians? What are the expressed

requirements in this regard?

(2) What are the objectives corresponding to such requirements in terms of mathematical mod-

eling?

(3) In what way medical data, particularly in this era of open medical data proliferation, makes

data mining more challenging?

(4) To what extent are the current data mining techniques able to satisfy the clinicians’ needs

and to handle the particular nature of medical data simultaneously?

In answering these questions, we are led to describe a comprehensive expert-aware approach

which stands out from the existing literature, through three main contributions exposed below.

• Because of the limited eﬀectiveness of some models, Karpatne et al. (2017) push for a theory-

guided data science. Such DM models are grounded in theoretical bases, in the domains of

Physics and Chemistry mainly. In the context of medical diagnosis, we can adopt a similar

approach, not guided by theory, but rather by the experts’ domain knowledge. Our paper

lays the bases for such an approach, in building a kind of bridge between both the medical

and data mining domains.

• We not only express that the issue of diagnosis aid is of a particular nature, we also propose

the translation of the associated speciﬁcs into modeling goals. Indeed, most of the papers

that have interest on the speciﬁcs of the medical domain have a wide scope, and are thus

not speciﬁcally focused on diagnosis, but also on prognosis and monitoring notably, which

involves that modeling is not discussed with enough depth (Bellazzi and Zupan, 2008; Cios

and Moore, 2002; Lavraˇc, 1999). Besides, we bring a more recent point of view compared

to the papers that speciﬁcally addressed aided medical diagnosis (Wagholikar et al., 2012;

Kononenko, 2001).

• We do not provide an overview of DM techniques and the related works; this was widely

proposed in previous surveys (Kalantari et al. 2018; Kourou et al. 2015; Esfandiari et al. 2014;

Wagholikar et al. 2012; Yoo et al. 2012). We rather question the existing DM techniques,

given the modeling goals raised following the understanding of the problem and data. This

allows us to raise some solid future research directions.

Predicted as  N P

Negative (N) TN FP

Positive (P) FN TP

Figure 2: Confusion matrix

The paper is organized as follows. In section 2, we expose the materials we considered to

structure and make our survey. The results are presented in section 3 and discussed in section 4.

Finally, we conclude this report in section 5.

2. Materials

2.1. Terminology

Medical diagnosis is the result of a challenging task which consists of collecting and conciliating

diﬀerent information (Donner-Banzhoﬀ et al., 2017; Hommersom and Lucas, 2016; Miller, 2016).

The latter include the symptoms (subjective data) and the signs (objective data) of the trouble

provided by clinical examinations and laboratory tests. In quest of explanations for these symptoms

and signs, the clinicians come to the conclusion of the existence/absence of a trouble, i.e. the

diagnosis.

A test is one among other elements that motivates a diagnosis (Gordis, 2014; Cios and Moore,

2002). The predictions of a clinical test are of several types. A patient with (respectively without)

the disease D predicted as such is designated as true positive (resp. true negative). In case of wrong

predictions, the patients are false positives or false negatives respectively. Let T P (resp. T N)

denote the number of True Positives (resp. True Negatives) and F P (resp. F N) the number of

False Positives (resp. False Negatives); these quantities are usually exposed in a matrix of confusion

(see Figure 2) (Witten et al., 2005). Diﬀerent scalar metrics are computed from T P , T N, F P and

F N to assess the performance of clinical tests; they are exposed in Table 1 (Lalkhen and McCluskey,

2008; Akobeng, 2007a,b). Let us note that positive and negative predictive values depend on the

prevalence of the disease (Akobeng, 2007a): they are easily deduced from the knowledge of sensitivity

and speciﬁcity, which are free from such an inﬂuence.

When several tests are required to check the presence of a medical condition, these tests may

be assessed globally in terms of net sensitivity and net speciﬁcity. The values of these indicators

depend on the way in which the tests were administered, i.e. sequentially or simultaneously (Gordis,

2014). Figures 3 and 4 present the mechanisms of sequential and parallel testing. For illustration

评论收藏

内容反馈

猫头虎

粉丝: 24w+
资源: 412

2022年医疗数据挖掘论文题材.zip

离群点分析与异常检测python源码(数据挖掘作业).zip

基于Q-Learning算法实现的论文推荐系统python源码(带数据和论文).zip

宠物小精灵数据挖掘-pokemon.zip

基于随机森林实现的心脏病分类预测python源码(数据挖掘作业).zip

2022年国赛优秀论文B086.zip

2022年国赛优秀论文B035.zip

2022年国赛优秀论文A171.zip

2022年国赛优秀论文B030.zip

2022年国赛优秀论文A001.zip

2022年国赛优秀论文A022.zip

基于python实现电信客户流失预测与分析源码(数据挖掘).zip

曙光数据挖掘实验代码.zip

中文医疗问答数据集.zip 非常实用的数据集

JAVA002打飞机游戏设计（程序+论文）.zip

2022美赛A题O奖论文.zip

yolo，yolov2和yolov3的论文原文.zip

Python数据挖掘实战课件.zip

无水印-2022年美赛特等奖论文集.zip

数据挖掘1.zip.zip

解决win7win8win10装4.8-3.5的.Net framework3.5安装失败问题 附带安装文档

谷歌浏览器axure扩展程序

时序图画图工具-TimeGen3.2安装包

百度、高德、腾讯、天地图、谷歌、必应、MapBox等地图金字塔切图工具 MapCutter 3.11.2

大唐杯习题合集-历年真题模拟题

zotero-pdf-translate-1.0.24（2023年7月10日）

姓名变为拼音.bas

typora1.8.10相关资源补丁

CiteSpace5.6.R2，目前最稳定的版本，改时间就可以用

小米盒子3，MDZ-16-AA，系统镜像文件1.4.16d.full.img

最新资源

解决win7win8win10装4.8-3.5的.Net framework3.5安装失败问题附带安装文档