E-MEDI: A WEB-BASED E-TRAINING PLATFORM

FOR BREAST IMAGING

I. Pratikakis, V. Virvilis, D. Kosmopoulos, S. Perantonis

IIT / NCSR “Demokritos”, Athens, Greece

A. Damianakis, D. Tsatsos

Conceptum SA,12 Heyden str., 10434 Athens, Greece

Keywords: e-Training, Breast imaging, web-based learning.

Abstract: This paper concerns a Web-based e-Training platform that is dedicated to multimodal breast imaging. The

assets of this platform are summarised in the following : (i) the efficient representation of the curriculum

flow that will permit efficient training; (ii) efficient tagging of multimodal content appropriate for the

completion of realistic cases and (iii) ubiquitous accessibility and platform independence via a web-based

approach.

1 INTRODUCTION

The essence of a medical doctor’s learning

curriculum is the acquaintance with as many

individual clinical cases as possible since the rate of

success in diagnosis and treatment is directly

proportional to the amount of this accumulated

experience. This learning principle is directly applied

to the field of radiology, where intensive training

with medical images is required. Nowadays, the

training in radiology at a European level is quite

diverse both in curricula and quality due to the lack

of appropriate content (tutorials and case-based

learning material) as well as the lack of qualified

trainers.

Furthermore, there is an increasing demand for

recruiting personnel specialized in radiography

which is proportional to the increase of produced

imaging volumes. In Sweden, imaging volumes are

increasing by 2% to 5% per year. The U.K. and

Canada are both seeing demand for imaging growing

by 5% (ESS, 2000). The demand isn't likely to fall. It

is obvious that there is a gap which may be largely

attributed to the deficiencies of the classic training

methods described in the following.

The number and complexity of the radiographic

imaging modalities have increased dramatically over

time. The last decades have brought about

tremendous innovation in the field with magnetic

resonance imaging, mammography, electron beam

computing tomography, positron emission

tomography just to name a few. While the

multimodal imagery helps the doctor in the diagnosis

it also represents an administrative burden that

renders the work of radiologists more complex,

while the technical and IT skill demands become

increasing and the underpinning knowledge base also

expanding. The range of tasks radiologists have to

perform has increased too. These rapid developments

not only require a well-prepared work force but also

rapidly adapting training programs.

The importance of critical thinking and

professional judgement in professional practice is

obvious. Especially the medical doctors that work in

geographically isolated regions are not able to

contact experienced professionals and lack in

experience due to the fact that the cases they handle

are very few compared to the ones handled in urban

hospitals. The traditional training programmes are

not applicable in these cases due to distance. The

principle of distance learning has been the

motivation factor for the increasing numbers of web

environments dedicated to learning / training

purposes, eg. (WebCT) (Blackboard) (TopClass)

(LotusLS).

In this paper, we present a Web-based training

environment that will facilitate accessibility and

support platform independence as well as will help to

bridge the gap between the availability of highly

338

Pratikakis I., Virvilis V., Kosmopoulos D., Perantonis S., Damianakis A. and Tsatsos D. (2007).

E-MEDI: A WEB-BASED E-TRAINING PLATFORM FOR BREAST IMAGING.

In Proceedings of the Third International Conference on Web Information Systems and Technologies - Society, e-Business and e-Government /

e-Learning, pages 338-345

DOI: 10.5220/0001281003380345

 SciTePress

trained/qualified radiology professionals and the

current needs in Europe for medical doctor’s training

and lifelong learning curriculum.

Emphasis will be placed on (i) the efficient

representation of the curriculum flow that will permit

efficient training; (ii) the tagging of multimodal

content appropriate for the completion of realistic

cases; (iii) a web-based platform that permits

ubiquitous accessibility and platform independence.

Furthermore the actual tagging can be seen as a part

of a strict protocol that helps the doctors file their

data sorted in a hierarchical way by case, time and

modality.

In this work, the focus is on breast imaging, but

the results can be easily extended to imaging

modalities of other anatomical objects like brain,

heart, etc.

The paper will be structured as in the following.

At Section 2, we will review existing relevant

environments for which we will address their

advantages and pitfalls. Section 3 is dedicated to the

overall system architecture wherein a detailed

description of the major components will be given.

Finally, at Section 4, concluding remarks will be

drawn.

2 RELATED WORK

In the following, we review many of the existing e-

learning systems related to radiology with particular

emphasis to breast imaging.

In (Costaridou et al., 1998), the potential of

interactive multimedia and Internet technologies is

investigated with respect to the implementation of a

distance learning system in medical imaging. The

system is built according to a client-server

architecture, based on the Internet infrastructure,

composed of server nodes conceptually modelled as

World Wide Web (WWW) sites. Sites are

implemented by integration and customization of

available components. The system evolves around

network-delivered interactive multimedia courses

and network-based tutoring, which constitute its

main learning features. This potential has been

demonstrated by means of an implemented system,

validated with digital image processing content,

specifically image enhancement. Image enhancement

methods are theoretically described and applied on

mammograms. Emphasis is given in the interactive

presentation of the effects of algorithm parameters

on images. The system end-user access depends on

available bandwidth, so high speed access can be

achieved via LAN or local ISDN connections

. In this

system, the content from real clinical cases was not

supported by all appropriate steps for a decision

making from the medical doctor. Furthermore, it

focused on mammograms only, limiting breast

imaging in a single modality.

In the case of the MammoEd project

(http:\\www.mammoed.com) which was developed

by the University of Washington, the aim was to

provide interactive, comprehensive teaching cases

that could be easily accessed from any computer

connected to the Internet and to provide general

breast imaging education resource for radiology

residents, attending physicians, students, clinicians,

technologists, and patients. It was developed using

teaching cases from daily clinical practice organized

into a computerized database. The screen-film

images were scanned. The student was prompted to

click on the pertinent findings or to answer questions

regarding the images. Each click rendered more

images and questions with discussion of the correct

and incorrect answers and management issues. Links

were embedded to relate teaching files and

references. Although, MammoEd system deals with

clinical cases from the daily clinical practice

supported by corresponding images that the trainee

can manipulate with minimal interactivity, it does

not support multimodal content and does not permit

the trainee to exploit all steps required for the

diagnosis of particular cases.

In January 2004 the Bavarian statutory Health

Care Administration started recertification program

for quality assurance and quality improvement in

mammography reading (Riesmeir, 2004). The

participating physicians are required to read 50 cases

randomly selected from a larger collection. The

mammography films were digitized using a high-

quality CCD scanner (570 dpi, 4096 shades of grey)

to be viewed on an appropriate display workstation.

In addition to the workstation software, a ‘home

edition’ operating on a standard PC window was

developed, allowing physicians to practice the

procedure at home and get used to working with the

software. Based on this ‘home edition’ of the

software, a CD-ROM with 35 cases for training was

composed and distributed. Future releases of the

home edition with more training cases will be in

DVDs. A shortcoming of this effort is the lack of

distance-based learning making the corresponding

material restricted to being consulted at a particular

place and consequently at particular time periods.

EURORAD (www.eurorad.org) is an e-learning

initiative of the European Association of Radiology

(EAR) that was officially launched at ECR 2001. It

is the first and still the largest peer-reviewed

Radiology teaching files database on the Internet,

and offers free access to a wealth of medical

information and imaging data, whose accuracy and

quality have been validated by some of the most

E-MEDI: A WEB-BASED E-TRAINING PLATFORM FOR BREAST IMAGING

339

experienced Radiologists in Europe. In this

approach, although we have a web-based learning

environment, it cannot permit any interactive

activities with the content itself.

In conclusion, it is clear from the above

overview that a more comprehensive e-learning

system for training is needed that firstly, will

encapsulate all available breast imaging modalities

(mammography, ultrasound, MRI), because

contemporary medical students are taught to

diagnose from a variety of modalities instead of

individual image sources. This ensures decrease of

uncertainties in difficult cases. The structure of such

a multimodal e-learning system should be adaptable

and customizable for any anatomical part (in contrast

to a system that would involve different single

modalities per body organ. Secondly, it is required

that training should be based upon a system that will

address all corresponding cases using real clinical

examples constructed in a manner that will respect

the procedure flow used in a daily clinical practice.

Finally, it is imperative that such a system should be

available for an ubiquitous trainee (anywhere,

anytime, from any computer system), enabling

accessibility, sharing and interoperability. The

proposed e-MedI framework strives toward fulfilling

the above requirements, for which a detailed

description is given in the following Sections.

3 E-TRAINING FRAMEWORK

The e-MedI architecture as it is shown in Figure 1 is

based on a client-server 3-tier architecture that

consists of the following core elements: (i) Learning

Management System (LMS); (ii) Visual Authoring

Tool; (iii) Trainee’s interactive e-training

environment and (iv) Keyword-guided Clinical Case

search tool.

3.1 LMS

The LMS is by definition a complex administrative

system used to deliver electronic content in the form

of lessons and to organize people who are attend

these lessons. In the proposed framework, we have

used a proprietary LMS. Specifically, the Lotus LMS

was chosen for a multitude of reasons such as

portability, compatibility, standard compliance and

feature completeness. Lotus LMS is written in Java,

which ensures, up to a certain degree, multiplatform

portability. Lotus provides LMS for several popular

and enterprise class platforms, thus actively

supporting the generic aspect of the learning

platform. Furthermore, LMS supports several

databases, both free and commercially available, as

its storage backend.

For a public access system such as a virtual

school authentication is an essential feature.

However, modern computing often requires

authentication more than one time. This situation

quickly becomes cumbersome encouraging people to

by pass security using weak passwords, sharing

accounts etc. Lotus LMS supports Single Sign On

(SSO) which enables the user to be authenticated

only once in order to access any service offered.

Figure 1: e-MedI e-training platform architecture.

LMS scales its support from simple seminars

lessons up to curriculum lessons sets. Although e-

MedI does use currently only a small percentage of

the full LMS feature set, this setup allows future

expansion.

Finally, Lotus LMS does support SCORM

(Sharable Content Object Reference Model) (LTSC),

the widely accepted standard for e-lessons, thus

making possible to take advantage of the multitude

of SCORM tools available in the market and in the

Free Software world, and avoid a potential future

vendor lock-in.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

340

3.2 Visual Authoring Tool

The Visual Authoring Tool is written in Java in order

to maximize portability and to take advantage of

future and current technologies like Java Web Start

(JWS) which drive the remote execution

technologies. The Authoring Tool is a custom

software made to facilitate the creation of Test cases

for breast imaging, but it can be adapted to other

types of pathology.

The main motivation factor that guided the

Authoring Tool development has been that the

Doctor's (teacher's) time is extremely valuable. With

the Authoring Tool, an experienced author is allowed

to construct a simple test case in about a time period

that ranges from 3 minutes, for a simple case, to 10

minutes for the most complex one.

In order to achieve this level of productivity and

user friendliness the specifications of the Authoring

Tool were crafted by an iterative process. The key

point was that we collected a series of feedback from

a team of doctors from different countries, namely

Ireland, Belgium and Cyprus. In this way, experience

was gathered from users that are familiar from

different medical systems and procedures. This

knowledge is embodied into the Authoring Tool.

The Authoring Tool creates Test Cases that

consist of consecutive stages. Each stage is

associated with one or more visual content object

(2D, 3D images and 2D+t sequences) . Each visual

content object is the acquisition output of one of the

available modalities (Mammogram, Ultrasound,

MRI).

At program startup, the Authoring Tool presents

to the author the opportunity to select and continue

editing an already existing case. If the author selects

a new case then he is prompted to select available

visual content objects from each modality. The

author is required to locate and describe the findings

the student is expected to identify in the image. The

description stage is supported via selection menus

(combo boxes), and boolean check boxes whenever

needed (Figure 2).

When the author has finished with the

description that is related to a particular visual

content object eg. a 2D image, it is prompted for

Figure 2: An instance of the e-MedI Visual Authoring Tool environment.

E-MEDI: A WEB-BASED E-TRAINING PLATFORM FOR BREAST IMAGING

341

another one. When finishes all visual content objects,

it is prompted for a new stage. Stages are data points

in patient's time line. Upon finishing all the desired

stages, the author is prompted to fill a small text field

representing comment's about the specific patient in

free text form, such as age or other important clinical

data.

Since e-MedI is a multilingual system and

despite the fact the Authoring Tool is currently only

in English the resulting Test case is multilingual with

the exception of the free text comment mentioned

above which has to be entered for all currently

supported languages (English, French, Greek).

Finally, it is worth mentioning that the Authoring

Tool gives an essential assistance to the tutors by

supporting image filters such as inversion, blurring

and sharpening that are routinely used by the medical

doctors during an examination.

3.2.1 Training Flow

A novel feature of the e-MedI e-training platform

relies upon the creation of a training flow that

assimilates the real situation in a clinical

environment and guides the interaction between the

trainee and the system during training. The training

flow depends on the selected modality, as it is

displayed in both the image properties panels and the

associated findings panels. For example in a

Mammogram image it is not possible to detect

‘nipple retraction’, so this specific option does not

exist. This way the Author faces only the relevant

choices, thus accelerating the process of content

creation. Examples of the availability of findings

with respect to the modality can be shown at Figures

6-7.

In the following, the produced XML

document that depicts the training flow is discussed.

After the XML preamble (header) that states that the

encoding is Unicode, each XML file fully describes

a clinical case. For each case, a short free text with

the patient's history is available along with a case

follow up for future reference. The case follow up

can be associated with a small image for a better

overview without actually digging into the case.

The XML format described so far is like this :

<?xml version="1.0" encoding="UTF-8"?>

<case>

<img>

Since each case may have multiple stages these

are described in the data section of the XML file.

<data>

<stage>

<image>

...

<help>

closeCase represents the final diagnosis and the

suggested course of action.

For example:

<closeCase>breastComposition="heterogeneo

usly dense" score="4: suspicious

abnormality" followUp="biopsy"></closeCase>

Each stage may have multiple images and each

image may have more than one finding. In each stage

it is possible for the author to specify further action

(e.g. request ultrasound) and give little guidance to

the trainees in the form of popup windows (<help>

tag)

Each area of the image has only one finding.

<stage>

<image>

<area>

<image>

The Image node has the following tags :

<file>: image filename

<label>: window title

<display>: left, right or both breasts

<modality>: mammo, ultrasound or MRI.

In the area tag the ROI coordinates are stored

along with the finding's type e.g. calcification. The

associated finding tag holds the options that are

displayed in the above figures and they are modality

specific.

Finally, the program gives the opportunity to

classify a finding as implant in order to avoid further

evaluation.

3.3 Interactive e-Training

Environment

The presentation tier is the layer of user interaction.

Its focus is on efficient user interface design and

WEBIST 2007 - International Conference on Web Information Systems and Technologies

342

accessibility throughout the training process. The UI

tier can reside on the user's desktop, on an Intranet,

or on the Internet.

The proposed core aspects of e-MedI as

experienced by the trainee during interaction are

presented in Figure 3. This flowchart shows the core

interaction between the trainee and e-MedI for a

single case. Outside of this core, additional

functionality will exist to manage case selection,

presentation and entry of cases into e-MedI. For

example, additional functionality could be envisaged

to filter cases into categories (eg. ‘easy’ to

‘difficult’), alter the appearance of the user interface,

or provide the trainee with scores and tutorials.

The data are not be revealed in their entirety to

the trainee at the outset but in stages. A stage is the

data that is available to the trainee at any given point

between decision nodes in the decision tree. For

example at the first stage, only the patient data and

clinical breast exam results may be available. On

selecting the correct action for this history, the

images and/or test results associated with the next

stage are displayed. Note that several views (images

taken at different orientations ) will typically exist

for a single examination. Where more than one

image exists for a given examination, images may

need to be presented as thumbnails or equivalent.

The trainee will then click on regions of the image(s)

to identify suspect lesions. The chosen locations will

be crosschecked with the stored lesion coordinates

for that image.

The trainee then enters a report for the current

image using a report data entry table, which displays

a range of standard image features such as “mass”,

“mass margins”, “mass density” etc. Each feature

has standard descriptors. For example the standard

descriptors of the margins of a mass in a

mammogram are “circumscribed’, “microlobulated”,

“obscured”, “indistinct” or “speculated”. Each

descriptor will be associated with a checkbox in the

report entry table. If the trainee believes the margins

of the mass identified are best described by

“circumscribed”, then they will check the box

associated with this descriptor. Errors between the

report entered by the trainee and the stored report

can be highlighted. The trainee must then select the

next action required from a predefined list. For

example possible actions include “carry out

ultrasound” and “carry out spot compression

mammogram”, “score BiRads” or “Biopsy”. Actions

can be selected using an on-screen push button,

dropdown list or similar. Other than scoring BiRads,

any non-imaging action will in general close the

case. An important feature is that several different

image modalities may be available at the same

screen providing the trainee the opportunity to

diagnose simulating the real diagnosis procedures.

This can be shown via a snapshot of the interactive

e-training environment at Figure 4.

3.4 Keyword-based Clinical Case

Search Tool

The e-MedI database holds all available visual

content along with their corresponding descriptors.

The keyword-based clinical case search tool allows

searching through keywords with the aim to assist

the user in finding visual content instances eg. 2D

images that have specific features in order to further

improve training in a particular case diagnosis. For

example if the user has in a mammogram that may

have a ‘ruptured implant’ and wants to verify the

diagnosis, he/she can use this feature as keyword to

retrieve all relevant visual content.

Searching is realised by querying with different

granularity. An example is shown at Figure 5. First,

the user of the system selects what type of image

wants to search eg. Mammogram, Ultrasound or

MRI. After that, based on the selected modality, the

user selects among all the modality-specific

descriptors those that are needed for the particular

Figure 3: Core interaction between the trainee and e-MedI

for a single case.

E-MEDI: A WEB-BASED E-TRAINING PLATFORM FOR BREAST IMAGING

343

image the user searches for. Some descriptors are

further refined by clicking on the button refine next

to it.

4 CONCLUDING REMARKS

The proposed platform for e-training on breast

imaging draws its importance from the underlying

principle that a fruitful training process has to

assimilate and subsequently follow a training flow

which corresponds to a valid step-by-step process

followed in a real clinical routine setting. This

principle is realised via e-MedI through mainly an

authoring tool that supports the creation of the

realistic flow as well as an interactive environment

that is both user friendly and enables the

acquaintance of the trainee with the training flow

process. e-MedI can support an ubiquitous trainee

due to its web-based design providing the students

useable skills that they can directly apply in the real

world. In our immediate future plans, we will apply

to corresponding authorities all over Europe that

support Continuous Medical Education certificates to

consider e-MedI as the means of accreditation on

breast imaging.

Figure 4: A snapshot of the interactive e-training environment.

Figure 5: Example of a refined query.

WEBIST 2007 - International Conference on Web Information Systems and Technologies

344

ACKNOWLEDGEMENTS

This research was supported by the EC Leonardo da

Vinci programme “E-MedI-Virtual School on

Medical Imaging and E-Learning Framewrok”

under contract 2004-EL/2004/B/F/PP-148269.

REFERENCES

Costaridou, L., et al., 1998. Distance Learning in

Mammographic Digital Image Processing, The British

Journal of Radiology, 71(842):167-74.

Riesmeir, J., 2004. Experiences with a workstation

Prototype for softcopy Reading within the Bavarian

Mammography Re-certificating Program, Acad Radiol,

11: 407-418

LTSC, Learning Technologies Standardization Committee

( http://ltsc.ieee.org )

WebCT, ( http://www.webct.com )

Blackboard, ( http://www.blackboard.com )

TopClass, ( http://www.wbtsystems.com/ )

Lotus LS, ( http://www.lotus.com/home.nsf )

ESS, 2000. Employers skill survey. Case study: Health and

social care, Department of education and employment,

Nottingham, UK.

Diagnostic imaging, http://www.diagnosticimaging.com

Figure 6: Valid findings for MRI.

Figure 7: Valid findings for Ultrasound.

E-MEDI: A WEB-BASED E-TRAINING PLATFORM FOR BREAST IMAGING

345