Quality of fit of Stable Mass-Spring Models

Abstract

Shape models are used for many tasks in modern image analysis, e.g. segmentation, tracking, etc. Rarely, the quality of fit of adapted shape models is automatically determined to decide, whether the model fitting was successful. This paper develops a principal strategy to measure this quality of fit and defines a specific measure in the case of Stable Mass Spring Models, which are especially appropriate for that. The quality of fit measure is tested in two medical segmentation cases (left ventricles in SPECT data and lymph nodes in CT data) and has been shown to work well. Using application specific thresholds on the quality of fit, it could be automatically detected, whether a segmentation was successful or failed.

Complete fully automatic model-based segmentation of normal and pathological lymph nodes in CT data

“…This way, besides a good search result, a segmentation result as good as possible is ensured, because high qualities of fit are normally connected to good segmentation results (see [25]). Figure 9 shows the effect of this model grouping.…”

“…Furthermore, they are locally flexible w.r.t. their shape, which complies with the nature of medical structures, and allow for a direct measurement of their quality of fit [25]. In [26], the direct predecessors of Fig.…”

“…In [25], this information is efficiently computed for SMSMs, which are pointed out to be especially suited for the calculation of the quality of fit. These pieces of information are combined to a quality of fit function, which can take values between 0 (bad quality of fit) and 1 (perfect quality of fit).…”

“…The result is calculated as a value between 0 (not fitting) and 1 (very well fitting). In [25], it is explained how the quality of fit can be calculated for SMSM, which are used in our method. Furthermore it states that these models are especially suited for the calculation of such a quality of fit, which was one important reason to choose this model type for our segmentation task.…”

“…"Requirements". Furthermore, as the lymph node model is an SMSM, a method exists to efficiently Best results achieve the models of scaling factor 6, 7, and 8, which have an initial diameter of 18, 21, and 24 mm, respectively calculate its quality of fit (see [25]), which is another requirement according to Sect."Requirements". With a diameter of 3 mm, the lymph node model was significantly smaller than its target structures in order to be able to initialize it completely inside the target structure, which is necessary for a reliable segmentation, before it adapts its size by its dynamic adaption to a target lymph node.…”