Conclusions

In spite of the enormous research and development effort invested into finding satisfactory solutions during the past decades, the problem of medical image segmentation (as image segmentation in general) is still an unsolved problem today, and no single approach is able to successfully address the whole range of possible clinical problems. The basic reason for this rather disappointing status lies in the difficulties in representing and using the prior information in its full extent, which is necessary to successfully solve the underlying task in scene analysis and image interpretation.

While first results already clearly demonstrate the power of the model-based techniques, generic segmentation systems capable to analyze a broad range of radiological data even under severely pathological conditions cannot be expected in the near future. Currently available methods, like those discussed in this chapter, allow to work only within a very narrow, specialized problem domain and fundamental difficulties have to be expected if trying to establish more generic platforms. The practically justifiable number of examples in the training sets can cover only very limited variations of the anatomy and are usually applied to analyzing images without large pathological changes. It still needs a long way to go, before the computer representation and usage of the prior knowledge involved in the interpretation of radiological images can be represented and used by a computer in complexity which is sufficient to reasonably imitate the everyday work of an experienced clinical radiologist. Accordingly, in the near future only a well-balanced cooperation between computerized image analysis methods and a human operator will be able to efficiently address many clinically relevant segmentation problems. Better understanding of the perceptional and technical principles of man-machine interaction is therefore a fundamentally important research area which should now get significantly more attention than what it was getting in the past.

0 0

Post a comment