An interactive Web-based tool for multiscale physiological data visualization

Studies of cardiovascular pathophysiology require an understanding of long-term trends of disease development in monitored subjects, and ECG is the primary signal of interest in such studies. An effective analysis most often relies upon an iterative viewing of an interesting feature in a long-term derived trend (heart rate, for example) and subsequent location of the underlying ECG waveform for a finer-resolution visualization. Traditionally such an iterative process is time-consuming, labor-intensive, and inaccurate. We developed a novel 'point-and-click' technology that allows extremely efficient isolation of raw physiological waveforms of interest based on long-term trend plots. The paradigm is as follows: a three-paned window shows, in the uppermost pane, a long-term ECG averaged feature trend (e.g. heart rate), which is automatically created by pre-processing software. Each data point in the top pane consists of approximately ten minutes of averaged data. By clicking on any given point in this uppermost plot, the middle pane fills with a time-series plot of instantaneous data from which the above data point was derived. Clicking on any point in the middle pane (e.g. instantaneous heart rate) causes the bottom plot to show the signal waveform (e.g. ECG) corresponding to that moment in time. The technology we developed utilizes multiple open-source software packages, including SVG, CSS, and CGI with back-end C-compiled binaries and Perl scripts that generate the graphics dynamically. This interactive Web-based software tool has turned a process of manually converting and charting data, which required weeks of work and extensive use of paper charts, into an entirely automated, paperless process that generates results in minutes, allowing the user to analyze long-term data with unprecedented efficiency. It can be extended to handle multiple feature trends, such as ST level, QRS width, etc. or correlation plots (e.g. QRS width vs. heart rate), where regions of interest need visual verification, and may also be adapted to other similar physiological signals.