“Imaging in the age of medical bioinformatics”

Radiologists are often faced with information from a variety of sources: images from different modalities, clinical data, and patient medical history. In these diagnostic imaging tasks, humans and computers have disjoint skills. Humans are good at accepting information from multiple sources, but are poor at complex decisions that involve assimilating large amounts of data. Computers are able to solve complex problems and can handle very large amounts of data. However, computers often have difficulty in deriving the salient pieces of information from multiple data sources. For example, humans can in general more easily determine the correspondence of a lesion from multiple images and can more easily interpret written text. It is vitally important that we develop techniques to allow translation of disparate pieces of information into forms that are readily acceptable to computers. Many man-years have been invested into trying to solve this problem in many different fields and applications. With the growing number of large databases in medical science, techniques such as data mining can find new solutions and new avenues for research to improve the health status of humans. Images are a potential source for knowledge discovery, but in their current form (a string of pixels) they are not amenable to computer analysis. Further the dictated radiology report describes the contents of the image and holds the key to understanding the clinical findings contained in the image, but again this text report is not interpretable to the computer. It is imperative to develop techniques to extract pertinent information from imaging exams. This paper illustrate some of these concepts using breast imaging as an example.