An IFS-Based Similarity Measure to Index Electroencephalograms

Abstract

EEG is a very useful neurological diagnosis tool, inasmuch as the EEG exam is easy to perform and relatively cheap. However, it generates large amounts of data, not easily interpreted by a clinician. Several methods have been tried to automate the interpretation of EEG recordings. However, their results are hard to compare since they are tested on different datasets. This means a benchmark database of EEG data is required. However, for such a database to be useful, we have to solve the problem of retrieving information from the stored EEGs without having to tag each and every EEG sequence stored in the database (which can be a very time-consuming and error-prone process). In this paper, we present a similarity measure, based on iterated function systems, to index EEGs.

Hadoop for EEG Storage and Processing: A Feasibility Study

“…It goes to prove that a cluster of commodity hardware (15 machines with Dual core processors and only 7.8GB of RAM here) is better at processing complex data than a single highly specialized powerful server if the task is a series of (semi-)independent steps that can run in parallel. Hadoop has also been shown to be able to process a national scale amount of data with a 13 all features except nearest neighbor synchronization 14 36 testing 14 combinations at a time and 1 testing 7 combinations at a time 15 compared to the estimated upper and lower bounds for the Python job respectively 16 files obtained by extracting all eyes closed segments from the original EDF+ files and applying each of the 9 tested features on the extracted segments 15 Result of 37 successive jobs instead of only one job testing all 511 combinations 16 estimates based on data available quite small number of cluster machines. This is also a rather cheap solution: a cluster like the experimental one costs 10000 to 20000 euros i.e 1000-1500 euros per machine as compared to above 3000 euros per machine for the type of server used in the Python experiments.…”

“…fractal dimension in [14,16]) prior to it. Other approaches ( [17]) select, quantify, visualize some "relevant" EEG features through time and present them to a practitioner who then interprets them and their variations to derive conclusions on the EEG.…”

“…Hadoop also scales linearly with the size of classification input files 16 ( Figure 4(c)) and handles feature dimensions' increase better than Python (about 2 orders of magnitude faster than Python (Figures 4(a) and 4(b))).…”

Large scale mining and evidence combination to support medical diagnosis

“…fractal dimension in [14,52]) prior to it. Other approaches ( [41]) select, quantify, visualize some "relevant" EEG features through time and present them to a practitioner who then interprets them and their variations to derive conclusions on the EEG.…”

“…The fractal dimension separates normal sequences and other sequence types ( [52]) and normal EEGs and Alzheimer patients EEGs ( [14]). An extra feature, the nearest neighbour synchronization (mNNC) (defined in 6 is computed in the feature computation step (to measure scalability) but not used for classification.…”