Toward Complete Medical Image Understanding
Faculty of Information Science, Hiroshima City University, Hiroshima, 731-3194, Japan
Before the new discipline of “computational anatomy” emerged, medical image understanding is central to medical imaging research. Most research has been aimed at recognition of anatomical (and/or pathological) objects in medical images, that is, image segmentation. As introduced later in this book, an array of techniques has been developed. Medical image segmentation research is generally targeted to specific organs such as the brain, lung, and liver, while several attempts at simultaneous segmentation of multiple organs have been performed. Also in some approaches, segmentation results of other organs are utilized to yield more reliable extraction of main target. However, the rest of the image, outside of the organ(s) of interest, is neglected.
By contrast, “complete medical image understanding” involves the extraction of all the data from organs and tissues within given medical images. The motivation for complete understanding is based on maximization of information extraction from each clinical dataset. As radiologists are always required to read the entire image (for instance, detection of bone metastasis in the spine is expected while reading a
Fig. 1.15 Anatomical classification: organ level (macro) to cell level (micro) (Ref. )
chest X-ray), computational understanding also needs to cover the entire area while, at the same time, not devoting attention to harmless findings or “incidentalomas.”