Paper · S7555
CellSeg: a robust, pre-trained nucleus segmentation and pixel quantification software for highly multiplexed fluorescence images
CellSeg combines a pretrained nucleus-segmentation model with image scaling, stitching, mask expansion and pixel statistics. Historical benchmark scores and representative tissue examples have different provenance and settings; densely packed tissue remains difficult. A colorectal comparison reuses 140 samples from 35 people and provides author-described gating and correlation. Correlation does not establish exhaustive cell truth or universal clinical accuracy. Source-native metric definitions, count conflicts and caption errors are preserved.
Read the paper at ebi.ac.ukWhat we checked
All 141 supplied XML units, 45 references and 17 initially omitted abbreviation terms were actually read. Six native main JPEG figures, printed image tables, gate percentages, correlation values and inline mathematical expressions were inspected. The advertised scope contains no separate supplements; six GIF previews are duplicate renditions without reading credit. Journal publication is January 18, 2022; PMC release is the following day. Nolan is credited with resources and funding; Lee and Bedia lead methodology and software.
Supplied text and six figures read
How we check sources
Not checked yet
- The original main-layout PDF, historical correction or version audit, linked reference bodies, software tutorials, raw datasets and code execution remain unreviewed.
- Gating statistics and the population-correlation script are available upon request and were not received. A second-array compensation result is described without shown data. Exact unprinted scatter values and every individual cell contour remain unknown.
Source limitations
- The source calls a metric precision while its denominator includes false negatives. Text calls 3,717 nuclei validation where the figure assigns them to test and gives 4,152 for validation; figure cross-references also conflict. No silent relabeling was performed.
- Chosen model settings, reused datasets and author-described blinding limit the comparison. Patent, stock, equity and advisory conflicts remain reported; agency funding establishes no classified invitation, programme access or UAP evidence.
Cite this record
Michael Y. Lee, Jacob S. Bedia, Salil S. Bhate, Graham L. Barlow, Darci Phillips, Wendy J. Fantl, Garry P. Nolan, Christian M. Schürch. CellSeg: a robust, pre-trained nucleus segmentation and pixel quantification software for highly multiplexed fluorescence images. January 18, 2022. BioParity S7555. https://bioparity.io/sources/S7555Record data (JSON)