We demonstrate that the proportions of different functional tissue units (FTUs) within prostate tissue constitute a simple, biologically interpretable, and effective whole-slide image representation for clinically relevant prediction tasks, including biochemical recurrence risk estimation and ISUP grade prediction. Despite their simplicity and substantially lower computational requirements, these features achieve performance comparable to more complex and computationally intensive approaches, such as self-supervised learning–based foundation models and attention-based multiple instance learning (ABMIL) methods.
