Autoencoder - a new method for keeping data privacy when analyzing videos of patients with motor dysfunction (P4.001)

Objective: To test whether coded frame vectors of autoencoders contain relevant information to analyse videos of patient movements and preserve data privacy. Background: Clinical assessment of disability, especially motor dysfunction, is crucial to monitor patient’s disease. Video recordings of patient performances are used to increase reliability in severity ratings. These recordings would allow automated, quantitative disability assessments by machine learning algorithms. However, creation of these algorithms usually involves non-physicians, which can pose a data privacy challenge. Autoencoders condense videos to their essential information content that is sufficient for algorithm development but not human-readable. They consist of an encoder that creats the condensed “coded videos” so called “coded frame vectors”, and a decoder that reverts the coded videos to the original video. Coded frame vectors can be shared with collaborators who are not permitted to identify the patient’s identity, while the decoder can be used to check if the essential information from the video has been captured. Design/Methods: An autoencoder was trained on 2,230 videos spanning 9 standardized motor performances recorded in 5 different settings. 20 pre-rated videos were encoded to create coded videos, and decoded again. Coded videos were shown to 10 physicians to test whether they were human-readable, and decoded videos were rated by 2 physicians to test whether ratings were equivalent and thus the information contained in encoded data was sufficient. Results: We were able to condense patient videos to their machine-readable essential information content to enable machine learning in a privacy-protecting way. Conclusions: Creating coded frame vectors with autoencoders are privacy-preserving vehicles for the transmission of video frame data to non-medical collaborators and provide a similar level of security to normal encryption - assuming that the encoder and decoder are not shared by the encoding party. Study Supported by: Novartis Pharma AG Disclosure: Dr. D`Souza has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Marcus DSouza received travel support from Bayer AG and research support from the University Hospital Basel. Dr. Johnson has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Microsoft Research. Dr. Dorn has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Novartis Pharma AG. Dr. Van Munster has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Caspar Van Munster has received travel support from Novartis Pharma AG, Sanofi Genzyme and Teva Pharmaceuticals; and honoraria for lecturing and consulting from Novartis Pharma AG, Biogen-Idec and Merck Serono; and compensation for serving in a scientific. Dr. Diederich has nothing to disclose. Dr. Kamm has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Christian Kamm has received honoraria for lectures as well as research support from Biogen, Novartis, Almirall, Bayer Schweiz AG, Teva, Merck, Genzyme, Roche and the Swiss MS Society (SMSG). Dr. Steinheimer has nothing to disclose. Dr. Kravalis has nothing to disclose. Dr. Boisvert has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Novartis Pharma AG. Dr. Ormesher has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Microsoft Research. Dr. Walsh has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Novartis Pharma AG. Dr. Sellen has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Microsoft Research. Dr. Dahlke has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Employee of Novartis. Dr. Uitdehaag has received personal compensation for consulting, serving on a scientific advisory board, speaking, or other activities with Genzyme, Biogen Idec, TEVA, Merck Serono, Roche. Dr. Kappos has received research support from Bayer HealthCare Pharmaceuticals, Biogen, F. Hoffmann-La Roche Ltd and Genentech,Novartis, Research grants from: the European Union, Roche Research Foundation, Swiss Multiple Sclerosis Society and Swiss National Research Foundation.