Methodology of statistical image analysis by using normal cases

Learning Objectives Understanding of methodologies for statistical image analysis is very important to have effective results from various quantitative image analysis tools. This exhibit explains technical terms of statistics, image processing, and experimental schemes for statistical image analysis.

This exhibit consists of following five parts: [1. Statistics] The average and the standard deviation, normal distribution, and confidence interval, [2. Statistical test] Student-T test and significance levels, [3. Image processing] Image deformation and registration by using Affine transformation and thin-plate-spline, [4. Statistical model by using normal cases] Procedures to construct the normal model and computation of Z-score, and [5. Overview of latest study] Brain and torso FDG-PET scan images. In each content, the technical terms and important references/papers were explained to understand the experimental design for statistical image analysis. The part one and two illustrate the fundamental terms to explain the statistical models with graphics to understand the idea of commonly used parametrical statistical test with probability model. The part three and four explain a required image processing technique to construct the statistical shape model. Various combination of rigid and non-rigid image deformation approach took an important part in the model configuration. The Z-score calculation based on the statistical model was also explained. The part five shows an overview of current schemes of statistical image analysis of brain, and a new idea for torso FDG-PET scan images that we have developed as the reading assisting tools of cancer diagnosis. A new concept of diagnosis assistant tool by using handheld devices were also demonstrated.

Research Support This work is supported partly by a grant from Japan Government Grant-in-Aid for Scientific Research (C) 19500385