Using Assurance Cases to Guide Verification and Validation of Research Software

TL;DR

This paper demonstrates how Assurance Cases can effectively guide verification and validation activities in research software, using a medical imaging case study to illustrate the approach and its benefits.

Contribution

It introduces the use of Assurance Cases for structuring VnV efforts in research software, emphasizing artifact traceability and explicit responsibilities.

Findings

Assurance Cases help identify necessary evidence for software confidence.

Explicit distinction between software and user responsibilities improves validation.

Code is not the sole artifact for building confidence in research software.

Abstract

Research software engineers can use Assurance Cases (ACs) to guide Verification and Validation (VnV) efforts. An AC is a structured argument that a property like correctness holds. We illustrate how ACs can guide VnV activities via a case study of software for automatically extracting the 3D segmentation of the aorta from medical images of the chest. The AC argument suggests that the following evidence is required: comparison to a pseudo-oracle; traceability between requirements, design, code and tests; review of all artifacts by a domain expert with proper credentials; documentation of input assumptions; and a warning that only qualified people should use the software. The case study highlights that code is not the only artifact of interest for building confidence and that making an explicit distinction between software and user responsibilities is useful.