Digital breast tomosynthesis (DBT) is a recent medical imaging tool that increases accuracy and interpretability compared to traditional full-field digital mammogram (FFDM). However, DBT interpretation time is estimated to be twice longer than for FFDM, explainable by its 3D nature. Computer-aided diagnosis (CAD) systems can help radiologists in their diagnostic tasks and workload reduction. However, computation times and costs are important for CADs, thus facing the same challenge as health practitioners. This study addresses the problem concerning the processing of DBTs with high cancer detection rates while meeting the constraints of the clinical world. To this end, we propose a method relying on the slabbing approach which generates a set of 2D thick slices “slabs” that summarize a whole DBT volume. We propose a comprehensive benchmark on slabbing exploring several parameters such as slab thickness and overlap between slabs. Our method uses a fully 2D convolutional neural network (CNN) as a binary classifier, trained solely on FFDMs, exploiting the similarity between FFDMs and slabs. We report metrics on the two publicly available datasets containing DBTs: Breast Cancer Screening-DBT (BCS-DBT) and EA1141. This is the first study to explore DBTs of the EA1141 dataset, so we provide data strategy details and make it publicly available on GitHub ( https://github.com/racoon-z/dbt-slabbing ). We report breast-wise \(\textrm{AUC}_{\textrm{ROC}}\)  of 0.90 on both BCS-DBT validation and test subsets and 0.97 on EA1141. We achieve competitive specificities at 90% of sensitivity breast-wise with 0.84 and 0.79 on BCS validation and test respectively, while not training on DBTs.