Improving Probabilistic Interpretation of Medical Diagnoses with Multi-resolution Image Parameterization: A Case Study

Abstract

Clinicians strive to improve established diagnostic procedures, especially those that allow them to reach reliable early diagnoses. Diagnostics is frequently performed in a stepwise manner which consists of several consecutive tests (steps). The ultimate step in this process is often the gold standard reference method. In stepwise testing, results of each diagnostic test can be interpreted in a probabilistic manner by using prior (pre-test) probability and test characteristics (sensitivity and specificity). By using Bayes' formula on these quantities, the posterior (post-test) probability is calculated. If the post-test probability is sufficiently high (or low) to confirm (or exclude) the presence of a disease, diagnostic process is stopped. Otherwise, it proceeds to the next step in sequence. Our case study focuses on improving probabilistic interpretation of scintigraphic images obtained from the penultimate step in coronary artery disease diagnostics. We use automatic image parameterization on multiple resolutions, based on texture description with specialized association rules. Extracted image parameters are combined into more informative composite parameters by means of principle component analysis, and finally used to build automatic classifiers with machine learning methods. Experiments show that the proposed approach significantly increases the number of reliable diagnoses as compared to clinical results in terms.