Abstract: 20
Type: Hot Topic
Abstract Category: N/A
Automated quantification of MR images is a goal towards which we can only strive. The main steps are high quality image acquisition and quality control, image pre-processing, segmentation of the structures of interest and detection of changes.
Image acquisition: Maintaining consistent image contrast is critical for reproducible quantification. This generally requires using the same pulse sequences on the same scanner, consistent head positioning, and adequate quality control.
Image pre-processing: Scans are acquired with varying geometric distortion, and arbitrary grey scale intensities. Gross distortion may be partially corrected on the scanner, but additional correction is usually necessary. As a first step, the brain is usually extracted from the skull and image background. Image intensity inhomogeneity then can be corrected by factoring out slowly varying biases of intensity across the brain. Arbitrary grey scale values then can be normalized so that tissue types have consistent image intensities across scans. Longitudinal pre-processing is superior to cross-sectional pre-processing for detecting changes over time.
Image processing: After image pre-processing, normal and abnormal tissues of interest can be identified (segmented), e.g., grey matter, white matter, lesions, etc. Focal and diffuse pathology in MS complicates tissue segmentation by altering tissue intensity profiles. Longitudinal image processing is preferable to detect change over time.
MS lesion segmentation and the detection of new/enlarging lesions is challenging. Lesions vary in intensity, size and shape. Gd-enhancing lesions appear on a background of enhancement of normal structures that dominate the image. New T2 lesions may be small, and may enlarge or become confluent with pre-existing lesions. Independently segmenting lesions on successive scans leads to poor characterization of lesion change due to segmentation variability. Difference images provide better visualization, but are complicated by intensity differences due to artifact and local misregistration. Sophisticated analytical methods are required to deal with this.
Many software approaches have been developed to quantify brain volume & volume change. The most popular in the MS field has been SIENAX and SIENA. Howver, there is increasing movement to the use of other longitudinal methods such as Jacobian integration.. Changes in the intensity of normal-appearing tissues complicate these measurements.
Disclosure: Dr. Arnold reports consultant fees and/or grants from Acorda, Adelphi, Alkermes, Biogen, Celgene, Frequency Therapeutics, Genentech, Genzyme, Hoffman LaRoche, Immune Tolerance Network, Immunotec, MedDay Merck-Serono, Novartis, Pfizer, Receptos, Roche, Sanofi-Aventis, Canadian Institutes of Health Research, MS Society of Canada, International Progressive MS Alliance, and an equity interest in NeuroRx Research.
Abstract: 20
Type: Hot Topic
Abstract Category: N/A
Automated quantification of MR images is a goal towards which we can only strive. The main steps are high quality image acquisition and quality control, image pre-processing, segmentation of the structures of interest and detection of changes.
Image acquisition: Maintaining consistent image contrast is critical for reproducible quantification. This generally requires using the same pulse sequences on the same scanner, consistent head positioning, and adequate quality control.
Image pre-processing: Scans are acquired with varying geometric distortion, and arbitrary grey scale intensities. Gross distortion may be partially corrected on the scanner, but additional correction is usually necessary. As a first step, the brain is usually extracted from the skull and image background. Image intensity inhomogeneity then can be corrected by factoring out slowly varying biases of intensity across the brain. Arbitrary grey scale values then can be normalized so that tissue types have consistent image intensities across scans. Longitudinal pre-processing is superior to cross-sectional pre-processing for detecting changes over time.
Image processing: After image pre-processing, normal and abnormal tissues of interest can be identified (segmented), e.g., grey matter, white matter, lesions, etc. Focal and diffuse pathology in MS complicates tissue segmentation by altering tissue intensity profiles. Longitudinal image processing is preferable to detect change over time.
MS lesion segmentation and the detection of new/enlarging lesions is challenging. Lesions vary in intensity, size and shape. Gd-enhancing lesions appear on a background of enhancement of normal structures that dominate the image. New T2 lesions may be small, and may enlarge or become confluent with pre-existing lesions. Independently segmenting lesions on successive scans leads to poor characterization of lesion change due to segmentation variability. Difference images provide better visualization, but are complicated by intensity differences due to artifact and local misregistration. Sophisticated analytical methods are required to deal with this.
Many software approaches have been developed to quantify brain volume & volume change. The most popular in the MS field has been SIENAX and SIENA. Howver, there is increasing movement to the use of other longitudinal methods such as Jacobian integration.. Changes in the intensity of normal-appearing tissues complicate these measurements.
Disclosure: Dr. Arnold reports consultant fees and/or grants from Acorda, Adelphi, Alkermes, Biogen, Celgene, Frequency Therapeutics, Genentech, Genzyme, Hoffman LaRoche, Immune Tolerance Network, Immunotec, MedDay Merck-Serono, Novartis, Pfizer, Receptos, Roche, Sanofi-Aventis, Canadian Institutes of Health Research, MS Society of Canada, International Progressive MS Alliance, and an equity interest in NeuroRx Research.