Disclosure:<\/strong> We created and published this dataset.<\/p>\n New Open-Access Benchmark: Hierarchical segmentation of vascular fibrosis in computational pathology.<\/p>\n The Core Challenge:<\/strong> Dataset Specifications:<\/strong> \ud83d\udcc4 Read the Data Descriptor: https:\/\/doi.org\/10.1038\/s41597-026-08214-y<\/a> submitted by \/u\/Logical-Internet-395 <\/a>
Segmenting the vascular wall is well-defined (baseline UNet DSC \u2248 0.80; inter-expert DSC \u2248 0.95). However, segmenting the intramural fibrotic microstructure remains an open problem. Due to its diffuse and heterogeneous nature, our baseline MONAI UNet achieves a DSC of only ~0.12 on fibrosis, while human inter-expert agreement averages ~0.61 DSC. <\/p>\n
\ud83d\udd2c Scale: 705 high-resolution micrographs (1534\u00d7780 px, 0.252 \u03bcm\/px), Picro-Mallory stain.
\ud83c\udff7\ufe0f Annotations: ROI + dual independent expert masks (vascular wall + fibrosis).
\u26a0\ufe0f Hierarchical Constraint: Fibrosis masks must be strictly spatially contained within the vascular wall.
\ud83d\udee1\ufe0f Robust Benchmarking: No color normalization applied; native aspect ratios preserved; strict animal-level 5-fold CV splits provided to prevent data leakage.<\/p>\n
\ud83d\uddc2\ufe0f Access the Dataset: https:\/\/doi.org\/10.6084\/m9.figshare.31386748<\/a><\/p>\n<\/div>\n
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
