Purpose

With the facility introduced by advances in medical technology for doctors to acquire detailed datasets from their patients, the volume of available medical information is fast becoming an ocean that is leaving many doctors struggling to the surface for a breath of air. Information continues to be collected at a rate far too quickly to be absorbed and, worse yet, critical information is more likely to be overlooked. Minerva aims to be a visualization and navigation environment that bridges the gap between clinical research and information visualization to facilitate data ingestion and communication.

Goals

Provide users with a high level abstraction interface of the metadata associated with a temporal heterogeneous dataset (i.e. a dataset including images, numerical data, text documents, etc.), while also providing an interface to analyze details with appropriate visualizations within the same spatial context.
Assist users in visualizing correlations among differing data modalities, as well as in discovering/verifying new correlations.
Preserve and share user parameter settings and insights, ultimately to the point of extending the domain knowledge base.
Allow users to aggregate the strengths of different visualizations when used in combination, and to familiarize themselves with new visualization techniques.
Treat spatial and non-spatial data equivalently, in order to maximize the benefit offered by scientific and information visualization techniques.

Components

Holistic Navigation

We are developing a method by which medical doctors can efficiently navigate an expansive set of images that have been collected throughout the medical history of a given patient. We take advantage of the metadata stored in image headers and other external resources, such as associated medical reports and anatomical references. Measures of navigation will include correlations in time, spatial proximity, medical focus, and disease.

Context Layering

Much research has focused on CAD (Computer-Aided Diagnosis) algorithms to increase diagnosis accuracy and minimize the time spent analyzing large sets of images. Surprisingly, the methods by which this information is presented to doctors and incorporated with the patients medical records is relatively simplistic. We believe that there is potential to improve the visual presentation of CAD results with the patient's images.

Our first intuition is to generate an imaging filter, or context layer, to the image for each CAD result. These context layers can be applied to the original dataset to focus the regions specified by the CAD but retaining the context of the surrounding regions. In addition to CAD defined context layers, it may be of interest to generate context layers defined by general image processing features, such as edge detection or image gradients, or defined by similar images of different modalities.

High Dimensional Data Visualization

Bioinformatics is a new and vibrant field of study that has developed in response to the rapid influx of medical information. The high dimensional nature of medical information is a non-trivial problem when researchers attempt to analyze data or draw correlations. Can doctors use previous medical records and patient tests to discover trends in disease patterns, detect health risks, or clarify ambiguous diagnoses? We plan to incorporate a data visualization and exploration tool in Minerva to be used as both a research and diagnostic tool on these fronts.