Clinical Evaluation

Evaluation and clinical validation of image reconstruction algorithms is inherently difficult and sometimes unconvincing. There is a clear need for guidelines to evaluate reconstruction techniques and other image processing issues in emission tomography. A particular concern in clinical studies is the tendency to compare not only different algorithms but different approaches to processing, without effort to isolate the effects due to the reconstruction algorithm itself. Examples are the comparison of iterative algorithms including attenuation correction with analytic approaches without attenuation correction or the comparison of different iterative algorithms where the fundamental implementation differs (e.g. use of "blobs"86 rather than pixels in the projector). This simply adds to the confusion in interpreting results. A further common problem is the comparison of clinical images where the reconstruction algorithm results in different signal to noise properties, typically dependent on the number of iterations utilised. Evaluation for a range of parameters tends to provide more objective results where the trade-off in noise and signal (e.g. recovery coefficient) can be more meaningfully compared.

Most of the algorithms developed so far have been evaluated using either simulated or experimentally measured phantom studies, in addition to qualitative evaluation of clinical data. This has been extended more recently to objective assessment of image quality using Receiver Operating Characteristics (ROC) analysis based on human or computer observers,87 evaluation of the influence of reconstruction techniques on tracer kinetic k

k parameter estimation88 and voxel-based analysis in functional brain imaging

using statistical parametric mapping. ,

0 0

Post a comment