The practice of Radiology has been qualitative ever since the discipline was established with invention of x-rays in 1897.However it moving towards Quantitative Radiology rapidly. For QR to become practical, the problem of image segmentation must be solved so as to offer adequate levels of automation and accuracy for any body region. We started the body-wide AAR project in 2009 with the goal of developing concepts, algorithms, and software for localizing, delineating, and quantifying objects in all body regions, via most common tomographic image modalities (CT, MRI, PET/CT) in a manner that is independent of objects, body region, and modality. The project has made significant advances in the neck, thorax, abdomen, and pelvis. It has also engendered several fundamental advances including the ideas of body-wide fuzzy modeling, hierarchical organizations of objects, optimal hierarchies, optimal recognition, model-based fuzzy connected delineation, recognition in images denoting texture properties, AAR in PET/CT images, rapid prototyping for applications, and several AAR applications.
Examples of object localization by the AAR approach in neck, thorax, abdomen, and pelvis in CT, PET/CT and MRI images are shown in the figure. Overall AAR object localization accuracy of 1-3 voxels and boundary delineation accuracy of 1-2 voxels have been demonstrated.