Non-parametric bootstrap ensembles for detection of tumor lesions

Summary Quantitative neuroimaging techniques have become emerging technologies within clinical practice. In this paper, we survey a few clinical applications where quantification methods have received increasing attention, namely in the area of brain atrophy, lesion load computation and quantification of diffusion processes. We focus on the reliability and reproducibility of such methods and will use the example of quantitative diffusion tensor imaging to discuss methodological details. There, we show possible avenues for evaluating correctness and reliability. On the one hand, we show results from our novel hardware phantom experiments, where axonal fibres are emulated by synthetic industry fibres. On the other hand, we present a new framework for constructing software phantoms which can be used to evaluate, for example, the impact of partial volume effects in case of axonal loss as to be found in multiple sclerosis. Advantages and disadvantages as well as pitfalls of quantification and evaluation techniques are illustrated throughout the paper.

show abstract

“…Bootstrapping has been used, e.g. in diffusion tensor imaging (DTI) quantification (3–5), segmentation (6) as well as in tumour diagnostics (7).…”

Section: Introductionmentioning

confidence: 99%

“…Bootstrapping has been used, e.g. in diffusion tensor imaging (DTI) quantification (3–5), segmentation (6) as well as in tumour diagnostics (7). Patient vs. volunteer studies: Although quantification algorithms may achieve reliable results for test data sets from healthy volunteers, algorithms can fail for real patient data.…”

Section: Introductionmentioning

confidence: 99%