Quantitative basis for medical imaging analysis. Information theory and Bayesian inference

One of the main issues in radiology is image analysis and report. The two stages of this process are the extraction of the information from the image and the diagnostic exercise. This paper is a theoretical analysis of image interpretation from a quantitative perspective, using information theory and Bayesian inference, and relating it to the concepts of evidence based medicine. It is concluded that the amount of information is dependent on the previous state of knowledge and the acquaintance with the clinical history and complementary exams. The three determined stages of the diagnostic process are a priori, conditioned by the image and a posteriori analyses. The first stage is related to the degree of information and knowledge of the case. The second stage is dependent on the conditional probability for the presence of image signs of the disease and it is thus directly dependent on the sensitivity and specificity of the exam. The third stage is dependent of the previous stages, that in the simplest case can be expressed in function of the odds or likelihood ratios. These three stages allow to reach a decision about the usefulness of a given exam and they orient the questions that a given image may answer. They also orient the research on image analysis for an adequate decision making, minimizing diagnostic uncertainty.