Datasets:

Yfeng0216
/

PriOrGen_DATA

language:
  - zh
  - en
license: mit
pretty_name: PriorGen Data
tags:
  - pathology
  - whole-slide-image
  - tcga
  - cancer-classification
task_categories:
  - image-classification
  - text-generation
configs:
  - config_name: default
    data_files:
      - split: train
        path: feature_labels.csv
  - config_name: modality_uni
    data_files:
      - split: train
        path: viewer/feature_labels_modality_uni.csv
  - config_name: modality_plip
    data_files:
      - split: train
        path: viewer/feature_labels_modality_plip.csv
  - config_name: organ_acc
    data_files:
      - split: train
        path: viewer/feature_labels_organ_acc.csv
  - config_name: organ_brca
    data_files:
      - split: train
        path: viewer/feature_labels_organ_brca.csv
  - config_name: organ_chol
    data_files:
      - split: train
        path: viewer/feature_labels_organ_chol.csv
  - config_name: organ_coad
    data_files:
      - split: train
        path: viewer/feature_labels_organ_coad.csv
  - config_name: organ_dlbc
    data_files:
      - split: train
        path: viewer/feature_labels_organ_dlbc.csv
  - config_name: organ_kich
    data_files:
      - split: train
        path: viewer/feature_labels_organ_kich.csv
  - config_name: organ_kirc
    data_files:
      - split: train
        path: viewer/feature_labels_organ_kirc.csv
  - config_name: organ_kirp
    data_files:
      - split: train
        path: viewer/feature_labels_organ_kirp.csv
  - config_name: organ_lihc
    data_files:
      - split: train
        path: viewer/feature_labels_organ_lihc.csv
  - config_name: organ_luad
    data_files:
      - split: train
        path: viewer/feature_labels_organ_luad.csv
  - config_name: organ_lusc
    data_files:
      - split: train
        path: viewer/feature_labels_organ_lusc.csv
  - config_name: organ_meso
    data_files:
      - split: train
        path: viewer/feature_labels_organ_meso.csv
  - config_name: organ_stad
    data_files:
      - split: train
        path: viewer/feature_labels_organ_stad.csv
  - config_name: organ_thca
    data_files:
      - split: train
        path: viewer/feature_labels_organ_thca.csv
  - config_name: tcga_cases
    data_files:
      - split: train
        path: viewer/tcga_cases/train.csv
      - split: validation
        path: viewer/tcga_cases/validation.csv
      - split: test
        path: viewer/tcga_cases/test.csv

PriorGen_Data

PriorGen 使用的数据独立存放于此目录，代码项目位于 /mnt/datadisk0/PriorGen。数据按用途组织：

PriorGen_Data/
├── features/uni/       # UNI WSI patch features (.pt)
├── features/plip/      # PLIP WSI patch features (.pt)
├── annotations/organs/ # TCGA organ report annotations and images
├── splits/             # train/val/test CSV and label mappings
├── knowledge_banks/    # organ knowledge banks (.pt/.json)
├── anchors/            # PIB anchor features (.pt)
├── feature_labels.csv  # 每个 UNI/PLIP 特征文件与 TCGA 器官类别的对应关系
├── DATASET_CARD.md     # 数据集说明与统计
└── dataset_manifest.json

tcga_cases 配置将合并划分表拆成 Viewer 可选择的 train、validation、test 三个 split；每行展示 slide ID、case ID、数字标签和器官类别。原始宽表也保存在 source/splits_ALL_MERGED.csv。

Viewer 配置选择器提供 modality_uni、modality_plip 两个模态子集，以及 organ_acc 至 organ_thca 的器官子集；default 配置包含完整清单。每个配置仍使用 train split。

feature_labels.csv 的每行对应一个特征文件，包含文件名、相对路径、模态（UNI/PLIP）、TCGA case ID、数字标签、器官类别和标签状态。器官标签编号对应：ACC、BRCA、CHOL、COAD、DLBC、KICH、KIRC、KIRP、LIHC、LUAD、LUSC、MESO、STAD、THCA。

代码默认通过 PRIORGEN_DATA_ROOT=/mnt/datadisk0/PriorGen_Data 读取数据，也可以在 PriorGen/.env.example 中修改路径。