Author's Reply to "MoNuSAC2020: A Multi-Organ Nuclei Segmentation and Classification Challenge"

Area of Science:

• Medical Image Analysis
• Computer Vision
• Biomedical Imaging

Background:

• The MoNuSAC2020 dataset is a large, manually annotated resource for medical image segmentation.
• A challenge was organized based on this dataset at ISBI 2020, with results previously published.
• An analysis by Foucart et al. identified potential errors in the challenge's evaluation metrics.

Purpose of the Study:

• To respond to the analysis by Foucart et al. regarding the MoNuSAC2020 challenge evaluation.
• To correct identified errors in the challenge's computational metrics and result tables.
• To provide an errata for the previously published challenge findings.

Main Methods:

• Careful examination of the analysis provided by Foucart et al.
• Identification and correction of a bug in the segmentation performance metric computation code.
• Correction of an erroneous column-header swap in a results table.

Main Results:

• A small bug in the code and an erroneous column-header swap were identified and corrected.
• After fixing the bug, the challenge rankings remained largely unaffected.
• Two suggestions from Foucart et al. were deemed valuable for future consideration but not immediate implementation.

Conclusions:

• The errata corrects minor inaccuracies in the MoNuSAC2020 challenge results.
• The corrections ensure greater accuracy in evaluating medical image segmentation performance.
• Further refinements to evaluation metrics are suggested for future biomedical imaging challenges.