Validation in Medical Image Processing: Methodological Issues for Proper Quantification of Uncertainties

Abstract

Validation is now considered as a mandatory step in any development of new medical image processing methods, systems, or components. It allows studying performances and adequacy with clinical applications, and comparison with similar solutions. Being able to correctly answer these three issues requires proper design, performance, and report of validation studies and results. This could be made easier by 1) the use of standardized solutions and 2) the explicit formalization and description of validation study components when designing and reporting such studies. In this paper, we will identify the major components involved in a validation study of a medical image processing (MIP) method, embedded or not in a wider clinical system. Emphasis will be given on the study conditions, and especially on the validation data sets. Main freely available validation data sets usable for MIP validation will be listed and shortly described. Finally, we will outline the need for validating the validation study itself and explain the two main aspects it includes. Keywords: Medical image processing, Validation, Validation data sets.