S Armato¹*, E Sidky²*, X Pan³*, M Genzel⁴*, J Macdonald⁵*, M Marz⁶*, Y Mansour⁷*, R Heckel⁸*, M Mazurowski⁹*, (1) The University of Chicago, Chicago, IL, (2) University of Chicago, Chicago, IL, (3) University of Chicago, Chicago, IL, (4) Utrecht University, Berlin, DE, (5) Institut fur Mathematik, Berlin, DE, (6) Technische Universitat Berlin, Berlin, DE, (7) Technical University Of Munich, Munich, DE, (8) Technical University of Munich and Rice University, Munich, DE, (9) Duke University, Durham, NC

• 
  S Armato
• 
  E Sidky
• 
  X Pan
• 
  M Genzel
• 
  J Macdonald
• 
  Y Mansour
• 
  R Heckel
• 
  M Mazurowski

WE-A-TRACK 5-0 (Wednesday, 7/28/2021) 10:30 AM - 11:30 AM [Eastern Time (GMT-4)]

Advancement of imaging research requires years of effort from dedicated research groups around the world. Such groups, working independently, typically suffer from limited local resources in terms of patient data and access to the “ground truth” required to properly train and test their algorithms. When these research groups report their methods in the literature, it is difficult for the research community to compare the relative merits of different approaches, since performance can depend on factors such as database composition, image quality, “truth” definition, and performance evaluation metric. Grand challenges allow for a direct comparison of different algorithms designed for a specific task, with all algorithms following the same set of rules, operating on a common set of data, and being evaluated with a uniform performance assessment approach. Comparisons among the methods of participating groups can help identify approaches that are the most promising for a specific task. The AAPM Working Group on Grand Challenges (WGGC) was created to promote the conduct of grand challenges in medical imaging by (1) developing recommendations for hosting computational challenges through AAPM activities designed to assess or improve the use of medical imaging in diagnostic and/or therapeutic applications, (2) vetting proposals from groups that wish to host a challenge in conjunction with the Annual Meeting, (3) facilitating the execution of these challenges, and (4) generating ideas for challenges to advance the field. Two challenges sponsored by the WGGC were conducted in the months leading up to the 2021 AAPM Annual Meeting and provided a unique opportunity for participants to compare their algorithms with those of other groups in a structured, direct way. Overviews of the Deep Learning for Inverse Problems: Sparse-View Computed Tomography Image Reconstruction (DL-Sparse-View CT) Challenge and the SPIE-AAPM-NCI DAIR Digital Breast Tomosynthesis Lesion Detection (DBTex) Challenge along with the DAIR Digital Breast Tomosynthesis Lesion Detection Challenge, Part 2 (DBTex2) will be presented along with presentations from the top-performing groups from the DL-Sparse-View and DBTex2 Challenges.

Learning objectives:
1. To understand the role of grand challenges and public image datasets in medical imaging research.
2. To learn about the methods used by the top-performing participants in the DL-Sparse-View CT Challenge.
3. To learn about the methods used by the top-performing participants in the DBTex2 Challenge.

Funding Support, Disclosures, and Conflict of Interest: The DL-Sparse-View CT Challenge was supported by NIH grant numbers R01 EB026282 and R01 EB023968 and by a Grayson-Jockey Club research grant. The DBTex2 Challenge was supported by NIH grant number R01 EB021360.

Handouts

• 166-57536-15711646-175232-1014086591.pdf (Emil Sidky)
• 166-60301-15711646-182020-52438332.pdf (Martin Genzel)

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email