Search

ObjectiveTo analyse the remodelling and recovery of a relocated intravitreal injection (IVI) service with an adapted treatment regimen in a tertiary referral centre during a nationwide lockdown with initial cancellation of all non-emergency treatments caused by the COVID-19 pandemic.Methods and analysisFor this retrospective observational study at Hanusch Hospital, Vienna, between 16 March 2020 and 5 May 2020, we conducted an analysis of an appointment booking system based on prioritisation incorporating disease class, severity and fellow eye status by evidence-based impact on irreversible structural impairment and survey data from telephone interviews. Recapture time was defined as the time-to-discard the backlog of patients in need for treatment. Non-attendance was stratified as treatment refusal for personal reasons and non-attendance due to lockdown-related restrictions.ResultsOf the 1109 patients, 241 (21.7%) were considered as highly urgent, 269 (24.3%) as urgent, 402 (36.2%) as semiurgent and 197 (17.8%) as non-urgent. Recapture time was 15 days for highly urgent patients, 22 days for urgent patients, 43 days for semiurgent patients and 46 days for non-urgent patients. The proportion of patients who refused treatment due to personal reasons was 5.2%, with a mean age of 82.4 years; 29 patients (2.6%) could not attend due to lockdown-related restrictions.ConclusionBy streamlining treatment based on urgency as well as increasing the number of bilateral IVI, recapture time was fast. We could provide a safe treatment environment for healthcare professionals and patients after resetting the injection service outside of the hospital with increased levels of protection.

This paper presents an exact pruning algorithm with adaptive pruning interval for general dynamic neural networks (GDNN). GDNNs are artificial neural networks with internal dynamics. All layers have feedback connections with time delays to the same and to all other layers. The structure of the plant is unknown, so the identification process is started with a larger network architecture than necessary. During parameter optimization with the Levenberg- Marquardt (LM) algorithm irrelevant weights of the dynamic neural network are deleted in order to find a model for the plant as simple as possible. The weights to be pruned are found by direct evaluation of the training data within a sliding time window. The influence of pruning on the identification system depends on the network architecture at pruning time and the selected weight to be deleted. As the architecture of the model is changed drastically during the identification and pruning process, it is suggested to adapt the pruning interval online. Two system identification examples show the architecture selection ability of the proposed pruning approach., {"references":["Y. Le Cun, J. S. Denker, S. A. Solla, \"Optimal Brain Damage,\" in\nD. S. Touretzky, \"Advances in Neural Information Processing Systems,\"\nMorgan Kaufmann, 1990, pp. 598-605.","B. Hassibi, D. G. Stork, G. J. Wolff, \"Optimal Brain Surgeon and\nGeneral Network Pruning,\" IEEE International Conference on Neural\nNetworks, vol. 1, pp. 293-299, April 1993.","R. Reed, \"Pruning Algorithms - A Survey,\" IEEE Transactions on\nNeural Networks, vol. 4, no. 5, pp. 740-747, September 1993.","M. Attik, L. Bougrain, F. Alexandre, \"Optimal Brain Surgeon Variants\nFor Feature Selection,\" in IEEE Proceedings of the International Joint\nConference on Neural Networks, pp. 1371-1374, 2004.","O. De Jes'us, M. Hagan, \"Backpropagation Algorithms Through Time\nfor a General Class of Recurrent Network,\" IEEE Int. Joint Conf. Neural\nNetwork, Washington, 2001, pp. 2638-2643.","O. De Jes'us, M. Hagan, \"Forward Perturbation Algorithm For a General\nClass of Recurrent Network,\" IEEE Int. Joint Conf. Neural Network,\nWashington, 2001, pp. 2626-2631.","O. De Jes'us, \"Training General Dynamic Neural Networks,\" Ph.D.\ndissertation, Oklahoma State University, Stillwater, OK, 2002.","O. De Jes'us, M. Hagan, \"Backpropagation Algorithms for a Broad Class\nof Dynamic Networks,\" IEEE Transactions on Neural Networks, vol. 18,\nno. 1, pp. 14-27, January 2007.","M. Hagan, B. M. Mohammed, \"Training Feedforward Networks with\nthe Marquardt Algorithm,\" IEEE Transactions on Neural Networks, vol.\n5, no. 6, pp. 989-993, November 1994.\n[10] L.S.H. Ngia, J. Sj¨oberg, \"Efficient Training of Neural Nets for Nonlinear\nAdaptive Filtering Using a Recursive Levenberg-Marquardt Algorithm,\"\nIEEE Transactions on Signal Processing, vol. 48, no. 7, pp. 1915-1927,\nJuly 2000.\n[11] P.J. Werbos, \"Backpropagation Through Time: What it is and how to to\nit,\" Proc. IEEE, vol. 78, no. 10, pp. 1550-1560, 1990.\n[12] R.J. Williams, D. Zipser, \"A Learning Algorithm for Continually Running\nFully Recurrent Neural Networks,\" Neural Computing, vol. 1, pp.\n270-280, 1989.\n[13] O. Nelles, Nonlinear System Identification. Berlin Heidelberg New York:\nSpringer-Verlag, 2001.\n[14] D. Schr¨oder, Elektrische Antriebe - Regelung von Antriebssystemen.\n2nd edn., Berlin Heidelberg New York: Springer-Verlag, 2001.\n[15] K. S. Narendra, K. Parthasarathy, \"Identification and Control of Dynamical\nSystems Using Neural Networks,\" IEEE Transactions on Neural\nNetworks, vol. 1, no. 1, pp. 4-27, November 1990."]}