11-05-2013, 03:28 PM
Mobile Medical Visual Information Retrieval
Mobile Medical Visual.pdf (Size: 781.19 KB / Downloads: 21)
Abstract
In this paper, we propose mobile access to peerreviewed
medical information based on textual search and contentbased
visual image retrieval. Web-based interfaces designed for
limited screen space were developed to query via web services
a medical information retrieval engine optimizing the amount of
data to be transferred in wireless form. Visual and textual retrieval
engines with state-of-the-art performance were integrated. Results
obtained show a good usability of the software. Future use in clinical
environments has the potential of increasing quality of patient
care through bedside access to the medical literature in context.
INTRODUCTION
HEALTH care decisions in hospital environments are often
refined during corridor discussions and while walking in
the hospital [1], [2]. Communication devices and access to information
outside clinicians’ offices are scarce, where only pagers
and wall telephones are usually available [3]. In most cases, a
need for access to information occurs in a given spatiotemporal
context, which has a limited shelf life. When the practitioner gets
back to the office, his attention is likely to switch to another task.
This is particularly true for the analysis of radiological images,
where visual memory cannot be summarized by a small set of
keywords that can be recalled when the physician regains access
to the related literature [4]. Ubiquitous access to peer-reviewed
medical information using mobile devices (i.e., smartphones)
and simplified interfaces has the potential to improve decision
quality and quickness [5]–[7]. Based on bedside consultations
of the literature, the clinician can iteratively ask questions to the
patient concerning anamnesis, physical exam or antecedents.
In-depth reading of the identified papers and further reasoning
can then be made at the time the physician gets back to his
office.
Mobile Medical Information Systems
Anytime, anywhere medical information access using mobile
devices has been proposed several times [1], [5], [6], [19]–[25].
A clear motivation from physicians was identified in [21], where
92% of 3482 physicians reported use of their personal digital
assistants (PDAs) multiple times per day to manage calendars,
access drug reference guides, and read medical journals. Physicians
in the military experienced the use of pocket personal
computers (PCs) also in a very positive way [23]. For emergency
situations outside of hospitals.
METHODS
Datasets and technologies used for web crawling as well as
document and image indexing are detailed in Sections II-A and
II-B [10], [31]. Solutions to make these techniques mobile are
explained in Sections II-C, II-D, II-E, and II-F.
Datasets
Articles from 24 journals from the online open-access publisher
BioMed Central7 in the fields of medical informatics and
medical imaging were used as the literature database of our system
(see Table II) available on the Internet. A second database
using articles from the two radiology journals Radiographics8
and Radiology9 were also indexed in the context of the Image-
CLEF10 benchmark. This second database is copyrighted and
thus not publicly available but on the other hand quantitative
performance measures of the tools are available [32].
Indexing the Medical Open Access Literature
Our retrieval engine MedSearch [10], [31] is based on existing
open-source tools for information retrieval. For text retrieval, the
open-source Java library Lucene11 was used. Lucene is flexible
and provides the possibility to index more than one field per
document, which enables searching in several data fields such
as text content, author name, and article title. For visual retrieval,
the GNU’s Not UNIX (GNUs) image finding tool (GIFT) was
chosen that has equally been used for almost ten years and has
shown to deliver stable visual search results. Implementation
details are given in Section III-A.
Web Services
Web services constitute an attractive way of establishing the
communication between the mobile interface and the literature
retrieval engine. Two web service technologies were investigated:
representational state transfer (REST) and simple object
access protocol (SOAP). REST is based on the hypertext transfer
protocol, whereas SOAP creates a new protocol for communication
with the client. This has the undesirable consequence of
increasing data exchange as it requires importing a library with
communication protocols. Since telecommunication operators
usually fix a quota for a data volume in a mobile contract, SOAP
web services may not be constituting an appropriate solution.
REST is simpler: the web service is called with a URL and
returns data to the client in XML format. Passing parameters to
the web service is straightforward, and the parameters may be
integrated directly into the URL.
DISCUSSIONS AND CONCLUSION
Medical visual information retrieval on mobile devices was
made possible with the MedSearch Mobile prototype described
in this paper. The initial MedSearch information retrieval engine
was successfully adapted to the various constraints imposed by
mobile devices. Enhanced ease of use of the interface and optimized
screen space were achieved using open-source libraries
for the development of mobile web applications.