Effective resource management is an important method for enhancing the handling of financial and critical safety aspects in modern hospitals. Especially in the operating room (OR) of the future potentials have to exploited to reduce costs of intensive and the risks of hazardous events [1–3]. Such events emerge for several reasons in the OR, like long idle times between sequenced surgical procedures or cancelled interventions due to poor coordination of staff and devices [4, 5].
A straightforward automated time and resource management system could be used to excellent effect to overcome these conflicts. Several approaches towards the real time prediction of phase and intervention time based on a wide range of surgical activities available online are described in literature [6, 7]. In this work we compared single activities of the surgeon to each other in terms of their suitability for predicting the remaining intervention time. The goal of this work is to clarify how a frequency based analogy can predict remaining intervention time. A novel approach for comparison of low level surgical activities is described here. An advantage of this method is that it works with just one single recordable surgical activity. The method was based on frequency domain analysis of time series for surgical procedures, which are usually complex and hard to predict. The periodogram as a nonparametric spectral estimation method was applied to evaluate the eligibility for time prediction of the surgical activities based on forecast the intervention time.
1.1 Related work
Several approaches addressed the real time prediction of phase and intervention time [6, 7]. In many other works methods are described to support OR management based on information available preoperatively, such as the type of intervention [8, 9]. These methods used Hidden Markov Models or Bayesian analysis to describe surgical procedures. Furthermore, classification and comparison of populations (patients, surgeons or systems) was described in  and  based on measures of statistical comparison like average and standard deviation. The disadvantage of these methods is that the bulk of data, like surgical activities or intraoperative anaesthesia, are necessary to forecast the intervention time. In this work we determined the single most suitable surgical activity to predict intervention time in real time. Obviously accurate real time recognition of surgical activities is required. Surgical activities can be recognized by medical and video data from endo-scopes, anaesthesia systems or other signals which are already used in surgical activity recognition systems [6, 12]. These methods are limited to a single surgery type and cannot be transferred directly to each surgery type though. Furthermore, the recent developments in the domain of the integrated and networked OR provided new sources for readily available data concerning surgical activities . In conclusion specific signals transmitted inside an OR can be used for automatic real time recording of activities of instruments like a coagulator or a mill.
The purpose of this work is to evaluate which of the activities conducted by a surgeon is most suitable to predict intervention time. Based on spectral estimation and feature extraction sequences surgical activities are compared to one another.
2 Material and methods
In this section we describe four important steps to compare surgical activities in terms of their suitability for intervention time prediction based on spectral estimation.
2.1 Transformation of surgical activities in time series
First off, a cornerstone to compare surgical activities of a specific surgery type is to have recordings of these activities. In our work we relied on recordings in the form of individual Surgical Process Models (iSPM). The iSPM consists of formalized atomic process steps of a surgery; whereby five different perspectives are considered: functional (de-scribes the action), operational (describes the used instruments or devices), spatial (describes the treated body part of the patient), behavioural (describes temporal information) and organizational (describes the executing person) . We concatenated the attributes of the first three perspectives to build an identifier and represented the identified activities’ status over the fourth perspective (recorded intervention time). This was done in the form of binary activity sequences over time x(t); x(t) had level 1 while the activity was being conducted and 0 otherwise. By doing so we generated 35 activity time series for one specific surgery type. For this work we focused on lumbar discectomies. Due to several dependencies inside our identifier, like for instance the action coagulate and the instrument coagulator usually being used in the same activity, we obtained 31 independent activity time series.
2.2 Estimation of the PSD
The second key part is the ascertainment of the similarity between several activities. In order to do this we estimated the PSD for every binary sequence. Based on the straightforward method of power spectrum estimation the so-called periodogram of the discrete activity sequences x[n] is estimated. The periodogram is a classical method and was developed by Schuster . Based on the first derivation of the Fourier Transformation of the activity sequence x[n] the periodogram can be yielded. Thus, the normal form of the spectrum X(Ω) is given by The estimation of the periodogram was done with Matlab2014b. The periodogram is often applied in practice for the purpose of spectral estimation; furthermore, it can be calculated in an uncomplicated manner. The size of segments was calculated by choosing the smallest known time interval of any considered activity time series. The smallest time interval helped in fulfilling the Nyquist criterion, too. To address errors caused by leakage, we used the Kaiser window, which was chosen empirically.
2.3 Feature extraction
The third step is comparing the PSDs of the activity time sequences to each other by the extraction of well-known statistical features. These features provide new periodicity information about the activity time sequences. Using these we can compare the surgeon’s activities. The following five features are calculated for every considered activity sequence: Total Power (is the sum of the intensity), Dominant Frequency (is the frequency of maximum power in the PSD.), Mean Frequency (is calculated as the sum of product of the power spectrum and the frequency divided by the total sum of the power spectrum.) and the Median Frequency (is the first frequency which exceeds half of the power of the spectrum.).
2.4 Comparison of surgical activities
Finally the frequency-based similarities of several surgical workflows of a specific surgery type were calculated with regard to their activity time sequences. For this purpose the surgical workflows of every single activity time series were chosen from those activity sequences which had the smallest deviation of the calculated features’ values to each other.
3 Experiments and results
Our four step study was performed based on forty recorded surgical workflows of lumbar discectomies. The four steps were further explained in section 2. The surgical work-flows used, were manual recordings from the ICCAS from 2007 and represent lumbar discectomies conducted by surgeons with various skill levels; the neurosurgical work-flows were divided into several surgical phases: approach to disc (preparation), discectomy and closure. The cross leave-one-out validation was started one minute after the cut time of a surgical procedure and the examined interval increased by one minute of the intervention time. Based on the prediction of the remaining intervention time error the several activities were compared. The maximum durations of the different phases were varied; the duration declined from the preparation phase to the final surgical phase.
In our evaluation study the predicted remaining intervention time error was calculated to investigate the feasibility of a frequency based activity comparison. The prediction error of activities which had the curettes as used instrument in the preparation phase, those which had the drill as used instrument in the discectomy phase and those with the drape as used instrument in the closure phase had the best outcomes. The activity irrigate had a very low prediction error of 21.45% in relation to the maximum possible error and had the smallest mean prediction error (26 min 33s). However, many activities such as those identified by being conducted with the surgeon’s right and left hand of the neurosurgical intervention exhibited high fluctuations which are depicted in the large standard deviation and asymmetric distribution.
Small standard deviation was seen for activities that were performed with cottonoids or a drill. Moreover, the mean prediction error showed a marked increase for many activities performed on the muscle or vertebra of the patient and for cutting in the discectomy phase. On the contrary, surgical activities such as those performed on the patient’s dura unveiled a decrease of prediction error as the intervention progressed.
4 Discussion and conclusion
In this work we investigated every surgical activity during interventions of a specific surgery type based on spectral estimation. The periodogram as nonparametric spectral estimation was calculated. To obtain the significance of every activity we considered the predicted intervention time error of every surgical activity.
The activity irrigate had the best outcome with a mean prediction error of 26 min 33 s. The needed activity information could be provided by an integrated and networked OR based on the signals of an irrigation pump. The drawback is that the ascertained significance is only valid for lumbar discectomies. Thus, a more general available activity would be more significant. The activities of the surgeon’s right and left hand are applicable in a wider range of surgery types. The mean prediction error for activities performed with the right hand was 41 min 39 s and for those done with the left hand 45 min 3 s. Furthermore a disadvantage is the wide variance of prediction error over intervention time. The high standard deviation can be inferred by the long time and by surgical workflow dependencies of the presented method. The influence of the time of the recording and the influence of the type of work-flow that was used have to be investigated more deeply in future works. Additionally further studies have to investigate methods of decreasing fluctuation. The presented method showed a novel approach to compare surgical activities with regard to their similarity. Hence, the presented method can be applied in an effective resource and time management system . The presented method exhibits feasibility for being integrated in assistance systems and thereby enhancing the daily routine inside the OR. Furthermore, the documentation during surgeries can be enhanced e.g. by automatically generated post-operative reports.
K. Cleary and A. Kinsella, “OR 2020: the operating room of the future,” Journal of laparoendoscopic & advanced surgical techniques. Part A, vol. 15, no. 5, pp. 495, 497-573, 2005. Google Scholar
H. U. Lemke and M. W. Vannier, “The operating room and the need for an IT infrastructure and standards,” Int J CARS, vol. 1, no. 3, pp. 117–121, 2006.Google Scholar
J. V. Sutherland, van den Heuvel, Willem-Jan, T. Ganous, M. M. Burton, and A. Kumar, “Towards an intelligent hospital environment: OR of the future,” Stud Health Technol Inform, vol. 118, pp. 278–312, 2005. Google Scholar
J. G. Wright, A. Roche, and A. E. Khoury, “Improving on-time surgical starts in an operating room,” Can J Surg, vol. 53, no. 3, pp. 167–170, 2010. Google Scholar
B. Denton, J. Viapiano, and A. Vogl, “Optimization of surgery sequencing and scheduling decisions under uncertainty,” Health Care Manag Sci, vol. 10, no. 1, pp. 13–24, 2007. Google Scholar
N. Padoy, T. Blum, S.-A. Ahmadi, H. Feussner, M.-O. Berger, and N. Navab, “Statistical modeling and recognition of surgical workflow,” Med Image Anal, vol. 16, no. 3, pp. 632–641, 2012. Google Scholar
S. Franke, J. Meixensberger, and T. Neumuth, “Intervention time prediction from surgical low-level tasks,” Journal of Biomedical Informatics, vol. 46, no. 1, pp. 152–159, 2013. Google Scholar
B. J. Ammori, M. Larvin, and M. J. McMahon, “Elective laparoscopic cholecystectomy: preoperative prediction of duration of surgery,” Surg Endosc, vol. 15, no. 3, pp. 297–300, 2001. Google Scholar
R. Alvarez, R. Bowry, and M. Carter, “Prediction of the time to complete a series of surgical cases to avoid cardiac operating room overutilization,” Can J Surg, vol. 57, no. 11, pp. 973–979, 2010. Google Scholar
L. Riffaud, T. Neumuth, X. Morandi, C. Trantakis, J. Meixensberger, O. Burgert, B. Trelhu, and P. Jannin, “Recording of surgical processes: a study comparing senior and junior neurosurgeons during lumbar disc herniation surgery,” Neuro-surgery, vol. 67, no. 2 Suppl Operative, pp. 325–332, 2010. Google Scholar
G. Forestier, F. Lalys, L. Riffaud, B. Trelhu, and P. Jannin, “Classification of surgical processes using dynamic time warping,” Journal of Biomedical Informatics, vol. 45, no. 2, pp. 255–264, 2012. Google Scholar
T. Blum, H. Feussner, and N. Navab, “Modeling and segmentation of surgical workflow from laparoscopic video,” Med Image Comput Comput Assist Interv, vol. 13, no. Pt 3, pp. 400–407, 2010. Google Scholar
Bohn S, Lindner D, Franke S, Neumuth T, “An interoperability architecture for networked medical devices and its application to neurosurgery,” in The 15th International Conference on Biomedical Engineering (ICBME 2013), Singapore, 2013. Google Scholar
T. Neumuth, F. Loebe, and P. Jannin, “Similarity metrics for surgical process models,” Artificial intelligence in medicine, vol. 54, no. 1, pp. 15–27, 2012. Google Scholar
A. Schuster, “On the investigation of hidden periodicities with application to a supposed 26 day period of meteorological phenomena,” J. Geophys. Res, vol. 3, no. 1, p. 13–13, 1898. Google Scholar
P. A. Lynn and W. Fuerst, Introductory digital signal processing with computer applications, 2nd ed. Chichester [u.a.]: Wiley, 1998.Google Scholar
C. Herfarth, “’Lean’ surgery through changes in surgical work flow,” The British journal of surgery, vol. 90, no. 5, pp. 513– 514, 2003. Google Scholar
About the article
Published Online: 2015-09-12
Published in Print: 2015-09-01
Conflict of interest: Authors state no conflict of interest. Material and Methods: Informed consent: Informed consent has been obtained from all individuals included in this study. Ethical approval: The research related to human use has been complied with all the relevant national regulations, institutional policies and in accordance the tenets of the Helsinki Declaration, and has been approved by the authors’ institutional review board or equivalent committee.