Nuclear Medicine and Artificial Intelligence: Best Practices for Algorithm Development

Abstract

The nuclear medicine field has seen a rapid expansion of academic and commercial interest in developing artificial intelligence (AI) algorithms. Users and developers can avoid some of the pitfalls of AI by recognizing and following best practices in AI algorithm development. In this article, recommendations on technical best practices for developing AI algorithms in nuclear medicine are provided, beginning with general recommendations and then continuing with descriptions of how one might practice these principles for specific topics within nuclear medicine. This report was produced by the AI Task Force of the Society of Nuclear Medicine and Molecular Imaging.

Keywords: algorithm; artificial intelligence; best practices; computer/PACS; research methods; statistics.

FIGURE 1.

Trend in publications on AI within nuclear medicine according to Scopus (Elsevier). Word cloud contains most commonly used terms in recent abstracts.

FIGURE 2.

AI applications spanning the gamut of nuclear medicine subspecialties.

FIGURE 3.

Pipeline for AI algorithm development together with key considerations of each stage of development.

FIGURE 4.

Annotation quality as function of different labeling techniques for diagnostic applications. This hierarchy does not imply how useful annotation method is (e.g., expert labels are often more useful than simulations because of limited realism of simulated data).

FIGURE 5.

Different approaches to cross validation, depending on dataset size and whether model selection is needed. Figure illustrates 5-fold cross validation without model selection/hyperparameter tuning (top), 5-fold cross validation with holdout test set (middle), and nested cross validation (5-fold outer loop, 4-fold inner loop) (bottom).