Generalizing the evaluation of medical image processing tools by use of Gabor patterns.

OBJECTIVE: Tools for medical image processing are usually evaluated by observers with radiological experience and with complex tasks. For easing evaluation of filtering and enhancement tools, the observer's task can be generalized. METHODS: By describing aspects of the MCS method (Mammographic Contrast Sensitivity) we illustrate issues of selecting a metric for assessing visual performance, the observer's task and the image material to be used, aiming at a generalization of the design of studies for the evaluation of medical image processing tools. Concerning the metric, we distinguish acuity from contrast sensitivity. With respect to the observer's task, we distinguish tasks of discrimination from those at a higher level of recognition. Finally, we show the advantage of using medical images for evaluating image processing tools by comparing the results for measurements on homogeneous background and mammographic images. RESULTS: The perceptual level of the observer's task and the complexity of the used image material influences the outcome of observer studies, particularly also from crowding effects. The design of a study should minimize the impact of the observer's experience on the outcome. This can be achieved by using non-anatomical, standardized perceptual targets like Gabor patterns, used in the context of medical images. CONCLUSIONS: Understanding the concepts of perception helps designing observer studies that are as complex as required, but at the same time as simple and general as possible. Performing an observer study may be simplified by a study design which does not require radiological experience of the observers, if the study aims at the evaluation of tools that shall support basic perception tasks, such as e.g. contrast enhancement.

Analysis of mice tumor models using dynamic MRI data and a dedicated software platform*.

PURPOSE: To implement a software platform (DynaVision) dedicated to analyze data from functional imaging of tumors with different mathematical approaches, and to test the software platform in pancreatic carcinoma xenografts in mice with severe combined immunodeficiency disease (SCID). MATERIALS AND METHODS: A software program was developed for extraction and visualization of tissue perfusion parameters from dynamic contrast-enhanced images. This includes regional parameter calculation from enhancement curves, parametric images (e. g., blood flow), animation, 3D visualization, two-compartment modeling, a mode for comparing different datasets (e. g., therapy monitoring), and motion correction. We analyzed xenograft tumors from two pancreatic carcinoma cell lines (BxPC3 and ASPC1) implanted in 14 SCID mice after injection of Gd-DTPA into the tail vein. These data were correlated with histopathological findings. RESULTS: Image analysis was completed in approximately 15 minutes per data set. The possibility of drawing and editing ROIs within the whole data set makes it easy to obtain quantitative data from the intensity-time curves. In one animal, motion artifacts reduced the image quality to a greater extent but data analysis was still possible after motion correction. Dynamic MRI of mice tumor models revealed a highly heterogeneous distribution of the contrast-enhancement curves and derived parameters, which correlated with differences in histopathology. ASPC1 tumors showed a more hypervascular type of curves with faster and higher signal enhancement rate (wash-in) and a faster signal decrease (wash-out). BXPC3 tumors showed a more hypovascular type with slower wash-in and wash-out. This correlated with the biological properties of the tumors. CONCLUSION: With the described software, it was possible to analyze tissue perfusion parameters in small xenograft tumor models in mice. Our data correlated with histopathological data, and the qualitative and quantitative perfusion parameters could distinguish two tumor entities with different growth characteristics.