Visualizing Heart Data Visualizing Heart Data of a living entity by - PDF document

What do researchers seek? What do researchers seek? � � To achieve a better understanding of the state To achieve a better understanding of the state Visualizing Heart Data Visualizing Heart Data of a living entity by analyzing time- -series data series data of a living entity by analyzing time taken from blood pressure taken from blood pressure from Pulse Intervals from Pulse Intervals � � Tools exist (e.g. Spectral analysis, Wavelet, Tools exist (e.g. Spectral analysis, Wavelet, etc.) etc.) By By � These tools are nonetheless hard to interpret: � These tools are nonetheless hard to interpret: Juan Gabriel Estrada Alvarez Juan Gabriel Estrada Alvarez – The high irregularity in the data set causes – The high irregularity in the data set causes “ “noise noise” ” to show up, possibly hiding the juicy stuff to show up, possibly hiding the juicy stuff Typical Spectrum Typical Spectrum What do researchers want? What do researchers want? � Clearly it is not so simple to infer things from � � To be able to look at the data in a way that � Clearly it is not so simple to infer things from To be able to look at the data in a way that something that looks like this: something that looks like this: is easier to interpret is easier to interpret � � To have a means of classification of heart To have a means of classification of heart data based on the state of the ‘ data based on the state of the ‘patient patient’ ’ � � As a consequence, diagnosis would become As a consequence, diagnosis would become easier, and diseases might be prevented by easier, and diseases might be prevented by early detection early detection The Proposed Solution The Proposed Solution The Proposed Solution The Proposed Solution � Clustering on the (derived) pulse � � The GUI is similar to that of � The GUI is similar to that of TimeSearcher TimeSearcher Clustering on the (derived) pulse interval data as an attempt to classify; interval data as an attempt to classify; Toolbar Area � A � A TimeSearcher TimeSearcher- -like application to visualize the like application to visualize the Series data; data; Cluster/Individual information View � � Query boxes would be useful in examining Query boxes would be useful in examining common features across clusters; common features across clusters; Cluster � Zoom boxes would allow detailed examination � Selection Zoom boxes would allow detailed examination Time-series View of individual time- of individual time -series. series. Query refinement sliders 1

What has been done What has been done The issues that make it hard The issues that make it hard � Contacted the authors of � Contacted the authors of 1. A typical series is roughly about 7,000 data 1. A typical series is roughly about 7,000 data TimeSearcher; TimeSearcher ; points points � � Established (tentatively) the clustering Established (tentatively) the clustering 2. 2. Original data contains corrupted points due Original data contains corrupted points due algorithm to be used: Normalized algorithm to be used: Normalized to monitoring machine calibration to monitoring machine calibration version of the RMSD (average version of the RMSD (average 3. Series do not all start at the same time! Series do not all start at the same time! 3. geometric distance); geometric distance); Expensive pre Expensive pre- -processing may be required. processing may be required. � � Partial GUI (based on Harry Partial GUI (based on Harry 4. 4. User feedback? User feedback? Hochheiser’ Hochheiser ’s s source code) source code) Possible solutions Possible solutions Possible solutions Possible solutions � � One can notice similarities at first sight on the spectra: One can notice similarities at first sight on the spectra: 1. Use neighbour averaging to represent 1. Use neighbour averaging to represent several data points in one single point several data points in one single point 2. Recover missing points by averaging the 2. Recover missing points by averaging the immediate neighbours. immediate neighbours. 3. Maybe there exists a representation that Maybe there exists a representation that 3. allows comparison independent of allows comparison independent of “ “starting starting” ” and and “ “ending ending” ” points. The points. The spectrum of each series is a candidate spectrum of each series is a candidate � � This is evidence that clustering is possible This is evidence that clustering is possible Possible solutions What has changed Possible solutions What has changed BEFORE NOW 4. 4. User feedback is definitely desirable. User feedback is definitely desirable. BEFORE NOW � � � Series and clusters would be Series and clusters would be � Averaging of data points will be Averaging of data points will be Will contact Bruce Van Vliet Vliet for this for this Will contact Bruce Van displayed with full detail displayed with full detail done done purpose purpose � � � � Cluster view would allow Cluster view would allow Cluster view allows switching to Cluster view allows switching to querying on clusters only querying on clusters only viewing all series in the clusters viewing all series in the clusters selected and vice selected and vice- -versa versa (querying on time series would (querying on time series would then be allowed) then be allowed) � � Allow zooming in cluster and � Allow zooming in cluster and � An extra window will display An extra window will display individual views individual views time series in full detail to allow time series in full detail to allow comparison with other series. comparison with other series. Only display where zoom will be Only display where zoom will be supported supported 2

What Next? What Next? � � Contact Bruce for user feedback Contact Bruce for user feedback � � Implement clustering (including pre Implement clustering (including pre- -processing) processing) � � Implement the display areas Implement the display areas � Integrate with the existing querying implementation of � Integrate with the existing querying implementation of TimeSearcher TimeSearcher � Implement detailed view in separate window with zoom � Implement detailed view in separate window with zoom capabilities capabilities � � Tune up the GUI Tune up the GUI � � Acknowledgements: Acknowledgements: – Harry – Harry Hochheiser Hochheiser for kindly providing the source code of for kindly providing the source code of TimeSearcher TimeSearcher – Bruce Van – Bruce Van Vliet Vliet for kindly providing the data set for kindly providing the data set 3