Abstract | ||
---|---|---|
Although many interactive segmentation methods exists, none can be considered a silver bullet for all clinical tasks. Moreover, incom- patible data representations prevent multiple algorithms from being com- bined as desired. We propose the GeoMap as a unied representation for segmentation results and illustrate how it facilitates the design of an integrated framework for interactive medical image analysis. Results show the high exibilit y and performance of the new framework. Currently, fully automatic segmentation of medical images is neither feasible nor desirable. Having a \user in the loop" is necessary from both a clinical and a le- gal point of view. Following the paradigm of interactive segmentation, a number of approaches were proposed which combine the cognitive abilities and medical experience of humans with the reproducable accuracy and computational power of machines. Such approaches dier in how they balance speed, ease-of-use, accu- racy, reliability and other design criteria. No single method achieves the optimal balance for all classes of images or at least for all clinically relevant regions of a single image. Therefore, combinations of several methods are required. Current toolkits (e.g. (4,11)) usually contain various segmentation algorithms, but oer only limited means to combine them on a single image. Ideally, a fully integrated tool environment would make it possible to i) switch to the most ap- propriate method depending on the local image content (e.g. employ edge and region detectors in the same image), and ii) reuse components of one method in another one (e.g. Canny's hysteresis thresholding within a watershed segmenta- tion). Such combinations are currently dicult because dieren t algorithms use incompatible data representations for almost all levels beyond the pixel matrix. To solve this problem, a unied data representation based on sound computer science principles is needed. Therefore, we link image analysis know-how with the ideas of abstract data types and modern generic programming techniques (1). As a rst result of our research program we propose the GeoMap, a new representation based on topological maps (5) which covers the requirements of a large number of algorithms. By using GeoMaps for intermediate and nal segmentation results, the combination of algorithms is made possible and, in fact, easy to realize. Our new approach can be implemented very ecien tly, achieving interactive response times even on a low-cost PC. |
Year | DOI | Venue |
---|---|---|
2004 | 10.1007/978-3-642-18536-6_13 | Bildverarbeitung für die Medizin |
Keywords | Field | DocType |
abstract data type,ease of use,image analysis,cognitive ability,data representation,generic programming | Silver bullet,Computer vision,Scale-space segmentation,Segmentation,Computer science,Segmentation-based object categorization,Image segmentation,Artificial intelligence | Conference |
Citations | PageRank | References |
2 | 0.39 | 5 |
Authors | ||
3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Hans Meine | 1 | 38 | 6.05 |
Ullrich Köthe | 2 | 2 | 0.39 |
H. Siegfried Stiehl | 3 | 516 | 67.10 |