Paper Menu >>
Journal Menu >>
Information Fusion for Process Acquisition in the Operating
Thomas Neumuth, Christian Meißner
Innovation Center Computer Assisted Surgery (ICCAS)University of Leipzig
Abstract—The recognition of surgical processes in the operating room is an emerging research field in medical engineering.
We present the design and implementation of a instrument localization system that is based on information fusion strategies to
enhance its recognition power. The system was implemented using RFID technology. It monitored the presence of surgical tools
in the interventional site and the instrument tray and combined the measured information by applying redundant, complementary,
and cooperative information fusion strategy to achieve a more comprehensive model of the current situation. An evaluation study
was performed that showed a correct classification rate of 97% for the system.
Keywords-surgical procedure, workflow, proces modeling, operating room, computer-assisted surgery, information fusion
Acquisition and modeling of surgical processes in the
operating room is an emerging research field in medical
engineering. The on-hand availability of surgical process
models enables several technologies and applications, such as
assessment of surgical strategies, evaluation of surgical assist
systems, or process optimization.
Technical applications like workflow management support
in the operating room rely on process information too. For
these applications, the workflow management system needs to
know the underlying process it has to support. Therefore, the
recognition of the process is an indispensable step.
One current research objective is the automatic recognition
of surgical activities or partial information of them. Several
approaches focused on the recognition of partial process
information from surgical processes. Main data sources were
the recognition of surgical gestures or tool in video data [2–6],
from kinematic data from telemanipulators or from virtual
environments . Other works emphasized the recognition of
surgical actions by using force/torque signatures  or
acceleration sensors .Additionally, eye movements of the
surgeon , the location of the OR staff , or the
interpretation of patient’s vital parameter  were used to
recover surgical activities.
We present the design, implementation, and evaluation of
an online instrument recognition system that uses RFID data
to identify information about surgical activities. Our design
uses different information fusion strategies to optimize the
recognition abilities of the system.
2. Information fusion strategies
The process of interrelation of data and information from
different sources is called information fusion [14–17]. For this
purpose, data are matched, correlated, and combined to create
an abstract, but also more appropriate and more precise view
of the measured object or scene.
Durrant-Whyte introduced a classification scheme that
distinguishes several information fusion strategies according
to sensor types that are used for acquisition. He differentiated
redundant, complementary, and cooperative information
fusion. An overview of the information fusion strategies is
presented in Table I.
B. Redundant information fusion
If redundant information fusion is applied, information is
acquired by sensors of a similar type. Here it is the objective
to compensate measurement errors by combining multiple
sensors. By applying this strategy, each sensor detects the
same measurement parameters of the same object
independently from other sensors. By following this strategy it
is possible to enhance robustness and the margin of error of
the overall system. An example for redundant information
fusion is the supervision of an area with different cameras.
Using several cameras decreases the chance that single spots
in the supervised area are hidden from one of the cameras.
C. Complementary information fusion
Complementary information fusion is performed by
combining different sensors with non-redundant information
about the measured objects. The sensors work independent
from each other and are applied to acquire a more complete
representation of the current situation. An example for
complementary information fusion is the supervision of
different areas with video cameras to obtain a more complete
D. Cooperative information fusion
Cooperative information fusion is applied to derive
information or data from several sensors of different types.
Open Journal of Applied Sciences
Supplement：2012 world Congress on Engineering and Technology
ht © 2012 SciRes.195
This is achieved by the combination of two independent
sensors that may concern different objects. This strategy
reduces general uncertainty and enhances the system’s
robustness by combining different information sources. An
example for the application of cooperative information fusion
is the combination of video data with different types of
information such as weight information.
TABLE I: INFORMATION FUSION STRATEGIES
Strategy Types of sensors T
pe of obtained
information fusion Same sensor type Same object
information fusion Same sensor type Different objects
information fusion Different sensor
type Same object
3. System Implementation
E. System design
We implemented a RFID-based instrument localization
system as a mock-up in our demonstrator operating room at
the University of Leipzig. The system consisted of oneRFID
readers (Sirit-InfinityTM 510 UHF)with four polarized
circular patch antennas. They were placed in the operating
room to detect the presence and absence of surgical
instruments in different recognition zones.
Antennas S1 and S2 were mounted at the operating table
and antennas S3 and S4 at the instrument tray (see Figure 1).
Two of the antennas, S1 and S3, were mounted in horizontal
direction. The other two readers, S2 and S4 were mounted in
vertical direction. All readers had an average distance between
one and two meters to the site and to the tray.
We equipped eight surgical tools with RFID tags to locate
their position either in the interventional site or the tray. These
instruments were standard surgical tools that were used during
functional endoscopic sinus surgery (FESS) in
otorhinolaryngology. A list of the instruments is given in
Table IIand the instruments with attached RFID tags are
shown inFigure 3.
Data from the antennas were processed by a standard
notebook. As described in the following evaluation section,
we simulated surgical work steps of FESS procedures that
were reenacted by actors to establish a realistic scenario. Since
the information about the surgical instrument represents only
partial information of the overall surgical process, we
completed the other information by human observation
according to the schema presented in . To model the
complete surgical process model, a human observer added
information about the current surgical action (e.g. cutting,
suctioning, etc.) or the treated anatomical structure
(cavitasnasi, cell. ethmoidales, etc.) and worked as a “sensor”
Figure 1:Setup of the RFID readers in the operating room
F. Information fusion for surgical process modeling
Four fusion sites were established to complete our
information fusion system for instrument recognition (see
Figure 2). The information fusion itself was performed
stepwise using different processing layers.
Redundant information fusion was realized in the fusion
sites F1 and F2. F1 joined the information from the horizontal
antenna S1 and the vertical antenna S2. Both antennas covered
the interventional site to detect the presence of a surgical tool
in the site. F2 joined the information from the horizontal
antenna S3 and the vertical antenna S4 from the instrument
tray. Both redundant fusions were implemented by a disjoint
fusion, meaning that if one of the antennas in each fusion site
detects a tag, the surgical instrument was registered as present
in the respective zone. By using this strategy it was possible to
deal noise originating from signal reflection or signal damping
by the hand of the surgeon that holds the instrument.
We combined the information originated by the fusion
sites F1 and F2 to realize complementary fusion in the fusion
site F3. Here the results of the instrument recognition in the
interventional site and the instrument tray were joined to
intensify the information about the presence of an instrument
in the site, which was the main interest for surgical process
modeling. Consequently, F3 detected an instrument in the site
if it was detected directly by F1 or it was explicitly detected as
absent from F2.
Finally, we joined the information of the presence of a
surgical instrument in the site with the additional information
about the surgical work step that was generated by the sensor
S5. This fusion was considered as cooperative and was the
final step to create the surgical process model.
ht © 2012 SciRes.
Figure 2: Information fusion setup for instrument localization
4. system evaluation Study
G. Study setup
To evaluate or information fusion system, we performed a
comprehensive study. The objectives of this study were
the verification of the function of the RFID-based instrument
localization system and the evaluation of the contribution of
each fusion stage to the overall application of surgical process
We simulated nine FESS procedures with approximately
650 different surgical work steps. The simulations were
performed by reenacting the procedures according to a
screenplay script by trained actors. To dissemble the patient,
we used a 3D-rapid prototyping model that was generated
from a CT scan of a real patient.
TABLE II: LIST OF INSTRUMENTS THAT WERE EQUIPPED WITH RFID-TAGS
No. Instrument with RFID ta
Blakesley forceps, angle
Blakesley forceps, straight
Suction tube, straight
Suction tube, angled
Figure 3: Surgical instruments equipped with RFID-tags
H. Study results
Our information fusion system performed a correct
classification of the instruments in the 650 surgical work steps
in the interventional site (F1) in 92% (sd=11%) and also in the
instrument tray (F2) in 92% (sd=10%) by mean.
Complementary fusion (F3) was performed correctly by 91%
(sd=11%) and cooperative fusion was performed correctly by
No significant different was found between the recognition
of the sensors S1 and S2. Sensor S3 at the tray performed
significantly better than sensor S4 (p<0.001). The direct
detection of the instrument presence by F1 was found more
beneficial to the fusion F3 than the detection of the absence of
an instrument in F2. Finally, the sensor S5 contributed
significantly to generate a correct surgical process model.
Automatic recovery of the surgical process is an important
method for different application concerning computer assisted
surgery in the operating room.
We presented an information fusion system that is able to
detect the localization of surgical instruments in the situs and
the instrument tray. Our system is a beneficial module for
automatic recognition of surgical work steps and can be part
of an overall measurement system that consists of a number of
subsystems that are able to recognize other aspects of surgical
work steps, such as the current action or the treated anatomical
structure. We introduced several information fusion strategies
for instrument recognition and showed the benefits of their
 T. Neumuth, P. Liebmann, P. Wiedemann, and J.
Meixensberger, “Surgical Workflow Management
ht © 2012 SciRes.197
Schemata for Cataract Procedures. Process Model-based
Design and Validation of Workflow Schemata,” Methods
of Information in Medicine, vol. 51, no. 4, May 2012.
 N. Padoy, T. Blum, S.-A. Ahmadi, H. Feussner, M.-O.
Berger, and N. Navab, “Statistical Modeling and
Recognition of Surgical Workflow,” Medical Image
Analysis, vol. 16, no.3, pp632-641, Apr 2012.
 G. Sudra, A. Becker, M. Braun, S. Speidel, B. P.
Mueller-Stich, and R. Dillmann, “Estimating similarity of
surgical situations with case-retrieval-nets,” Stud Health
Technol Inform, vol. 142, pp. 358–363, 2009.
 F. Lalys, L. Riffaud, X. Morandi, and P. Jannin,
“Automatic phases recognition in pituitary surgeries by
microscope images classification,” presented at the IPCAI
 L. Bouarfa, P. P. Jonker, and J. Dankelman, “Discovery
of high-level tasks in the operating room,” J Biomed
Inform, Jan. 2010.
 H. Lin and G. Hager, “User-independent models of
manipulation using video contextual cues,” presented at
the MICCAI, London, 2009.
 B. Varadarajan, C. Reiley, H. Lin, S. Khudanpur, and G.
Hager, “Data-derived models for segmentation with
application to surgical assessment and training,” Med
Image Comput Comput Assist Interv, vol. 12, no. Pt 1, pp.
 A. Darzi and S. Mackay, “Skills assessment of
surgeons,” Surgery, vol. 131, no. 2, pp. 121–124, Feb.
 J. Rosen, B. Hannaford, C. G. Richards, and M. N.
Sinanan, “Markov modeling of minimally invasive
surgery based on tool/tissue interaction and force/torque
signatures for evaluating surgical skills,” IEEE Trans
Biomed Eng, vol. 48, no. 5, pp. 579–591, May 2001.
 A. Ahmadi, N. Padoy, K. Rybachuk, H. Feußner, S.
Heining, and N. Navab, “Motif Discovery in OR Sensor
Data with Application to Surgical Workflow Analysis and
Activity Detection,” presented at the MICCAI Workshop
on Modeling and Monitoring of Computer Assisted
Interventions (M2CAI), London, 2010.
 A. James, D. Vieira, B. Lo, A. Darzi, and G.-Z. Yang,
“Eye-Gaze Driven Surgical Workflow Segmentation,” in
Proc. of Medical Image Computing and Computer-
Assisted Intervention (MICCAI 2007), vol. LNCS 4792,
N. Ayache, S. Ourselin, and A. Maeder, Eds. 2007, pp.
 A. Nara, K. Izumi, H. Iseki, T. Suzuki, K. Nambu, and Y.
Sakurai, “Surgical workflow analysis based on staff’s
trajectory patterns,” presented at the MICCAI Workshop
on Modeling and Monitoring of Computer Assisted
Interventions (M2CAI), London, 2010.
 Y. Xiao, P. Hu, H. Hu, D. Ho, F. Dexter, C. F.
Mackenzie, F. J. Seagull, and R. P. Dutton, “An algorithm
for processing vital sign monitoring data to remotely
identify operating room occupancy in real-time,” Anesth.
Analg, vol. 101, no. 3, pp. 823–829, table of contents, Sep.
 D. Hall and J. LLinas, “An Introduction to Multisensor
data fusion,” 1997, vol. 85(1), pp. 6–23.
 Xiong N. and Svensson P., “Multi-sensor management
for information fusion: issues and approaches,”
Information Fusion, vol. 3, pp. 163–186, Jun. 2002.
 J. Llinas, C. Bowman, G. Rogova, A. Steinberg, E. Waltz,
and F. White, “Revisiting the JDL Data Fusion Model II,”
IN P. SVENSSON AND J. SCHUBERT (EDS.),
PROCEEDINGS OF THE SEVENTH
INTERNATIONAL CONFERENCE ON
INFORMATION FUSION (FUSION 2004, vol. 2, p.
 M. M. Kokar, J. A. Tomasik, and J. Weyman,
“Formalizing classes of information fusion systems,”
Information Fusion, vol. 5, no. 3, pp. 189–202, Sep. 2004.
 H. F. Durrant-Whyte, “Sensor Models and Multisensor
Integration,” The International Journal of Robotics
Research, vol. 7, no. 6, pp. 97 –113, Dec. 1988.
 T. Neumuth, P. Jannin, G. Strauss, J. Meixensberger, and
O. Burgert, “Validation of knowledge acquisition for
surgical process models,” J Am Med Inform Assoc, vol.
16, no. 1, pp. 72–80, 2009.
 T. Neumuth and C. Meissner, “Online recognition of
surgical instruments by information fusion,” Int J Comput
Assist Radiol Surg, vol. 7, no. 2, pp. 297–304, Mar. 2012.
ht © 2012 SciRes.