Quantitative Imaging Biomarker and Noise Characteristics in CT

• Activity Number: 26
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Imaging
• Author(s): Hyun (Grace) Kim*+ and David Gjertson and Matthew Brown and Jonathan Goldin
• Companies: UCLA and UCLA School of Public Health and UCLA Computer Vision and Imaging Biomarker and UCLA Radiology
• Keywords: biomarker ; denoise ; image ; normalization ; classification ; machine learning

Abstract

In radiological imaging, the data set is high-dimensional with the spatial correlation. To analyze the data set, it is important to strategize the goal of modeling with known structure of anatomy. Additional challenges are the standardization before applying to clinical trial and research application. Different technical settings produce a different level of noise and affect to the quantitative measurement of imaging biomarker. In setting of biomarker evaluation, the essential factors are calibration, discrimination and accuracy in multicenter clinical trial. We present a new denoising approach with consideration of anatomy, as part of calibration. Classification model has developed using a machine learning algorithms from the denoised features. The result is compared radiologist score with classification model as discrimination and accuracy. We, then, show the improvement of imaging biomarker in the comparison of the treatment outcome in a clinical trial.