Clinical proof of concept of dynamic reconstruction of digital breast tomosynthesis

<i>Objective.</i>Motion artifacts in digital breast tomosynthesis (DBT) are concerning in clinical practice. Although their impact varies, correcting them is crucial to maintain diagnostic performance. This paper presents advances in dynamic reconstruction for DBT, based on a previously proposed phantom-validated dynamic reconstruction methodology. The main contributions include qualitatively and quantitatively validating the methodology on a 212-case clinical database, proving that real texture and more complex kinematics are not a limitation. The introduction of local dynamic reconstruction described here is a fundamental aspect of this success.<i>Approach.</i>Building on a validated dynamic reconstruction strategy-combining motion-corrected reconstruction and projection-based digital volume correlation - the focus is on designing clinically tailored kinematic bases and evaluating their performance in local or global reconstructions. Local bases are derived from an incompressibility constraint and initiate the exploration on clinically exploitable bases, while a finite-element mesh supports the full-field correction.<i>Main results.</i>Validation of local dynamic reconstruction showed 90.1%-97.5% of cases had reduced loss, with improvements ranging from 11.8% to 15.4%. Kinematic comparisons with manually tracked references showed improvements in 54.6%-75.8% of cases. A preliminary human reader study found a 77.2% preference for dynamic reconstruction in cases where readers agreed. Initial evaluation of global reconstruction showed a 4.2% loss decrease and significant improvement in clinical features.<i>Significance.</i>This paper adapts a dynamic reconstruction strategy for DBT to clinical settings, leveraging specific kinematic bases and validating the method on a 212-patient dataset. The approach improves analytical reconstruction metrics, including loss reduction and alignment with manual kinematic references. A human reader study further supports the method, indicating better image quality in dynamically reconstructed volumes, with sharper edges and fewer replication artifacts. Importantly, the global finite-element model provides the first known full-field 3D dynamic reconstruction in breast tomosynthesis.