Synopsis

Keywords: Transferable skills: Reproducible research, Musculoskeletal: Knee, Transferable skills: Software engineering

Deep learning (DL) has shown promise for a variety of applications in MRI from upstream acquisition and reconstruction to downstream image analysis. However, while new DL models for MRI are being developed and open-sourced at unprecedented rates, their translation into tools that can be used by research and clinical practitioners has been limited. In this talk, we will discuss the challenges, solutions and opportunities for building user-centric, DL-assisted tools for MSK MRI workflows.

MRI workflows are composed of many different stages (e.g. reconstruction, segmentation, quantitative mapping), some of which can be augmented with deep learning (DL). Thus, translating new DL methods into practice involves more work than just building new models. It also requires building systems that simplify using these models as part of end-to-end user workflows. These user-centric systems should facilitate integrating DL models into existing user workflows, simplify the process for validating new DL methods on relevant metrics, and help define standards for image analysis, which is critical for reproducibility. In this talk, we will examine existing challenges to integrating DL models into MRI workflows and how dedicated open-source solutions can address these challenges. We will showcase examples of how these solutions can help with standardization, repeatability and reproducibility in MSK image analysis.

Acknowledgements

No acknowledgement found.

References

No reference found.