(Sunday, 7/12/2020) [Eastern Time (GMT-4)]
Room: AAPM ePoster Library
Purpose: To present a metrological approach to improved voxel-to-voxel image comparison based on joint probability, PAB, and pointwise mutual information (PMI) that lessens the impact of intensity/geometric mismatches.
Methods: PAB and PMI are derived from a 2D intensity-histogram. The PAB is mapped back to the image pair’s overlap locations to form a “PAB_image”. “Mismatches” are associated with a band of low PAB values over a wide range of PMI values in “PMI-PAB Map”. The distribution of the PMI-PAB values is bound by the “perfect intensity-geometry” (PIG) curve, PAB = exp(-PMI). “Acceptable” matches lie between the “mismatched band” and the PIG curve. To improve comparison one can either remove mismatches using metrological analyses, or enhance the PAB by clustering the conventional gray levels (256) into fewer gray levels. The latter approach is applied to a pair of pre- and post-therapy ADC MRI images of a patient with malignant glioma to evaluate its efficacy.
Results: The joint probability, PAB, usually has a very low value as it is spread-out over the entire image matrix size, in particular the PAB for a tumor, for which the intensity is expected to be different between pre/post-treatment, is even lower. However, using a clustered 16 gray level PAB one can increase the overall PAB by 20 times. As a result changes within the tumor volume can be visualized in the clustered PAB image, but not in the conventional gray-level image.
Conclusion: PAB and PMI analysis help improve voxel-to-voxel image comparison. Based on this a clustered 16 gray level PMI-PAB map was used to reduce the impact of intensity/geometry mismatches. The improvement allows one to correlate intensity pairs over the pair’s geometric overlap to quantify the changes for pre and post therapy ADC changes.
Keywords
Image Correlation, Mutual Information, Noise
Taxonomy
IM- Multi-Modality Imaging Systems: General (Most aspects)
Contact Email
