PANDA-PLUS-Bench: Prostate Cancer Pathology Patches for Model Evaluation

PANDA-PLUS-Bench is a benchmark dataset for evaluating feature collapse in pathology foundation models. It contains expert-annotated prostate biopsy image patches from 9 whole slide images representing 9 unique patients, with pixel-level Gleason pattern annotations. The dataset was created by author dellacorte and last updated on December 16, 2025.

Use Cases

• Benchmarking pathology foundation models based on the described feature collapse evaluation task.
• Training models for Gleason pattern classification based on the four annotated classes (Benign, GP3, GP4, GP5).
• Evaluating model performance on expert-annotated, pixel-level ground truth data from prostate biopsy patches.

Strengths

• Contains pixel-level Gleason pattern annotations by experts.
• Includes patches from 9 whole slide images representing 9 unique patients.
• Patches are provided at a resolution of 224×224 pixels at 20× magnification.

Limitations

• Dataset size is limited to approximately 2,770 patches per augmentation condition.
• Column-level documentation is absent; field semantics must be inferred after download.
• Geographic and demographic coverage of the 9 patients is unknown, which may limit generalizability.

Provenance

Source

dellacorte on Hugging Face

Collection Method

Expert-annotated prostate biopsy patches extracted from whole slide images.

Freshness

Last updated 2025-12-16 02:04:44; freshness should be verified.