RA_config.yaml

# Paths
paths:
  input_directory: "C:/Users/Amaya/Documents/PhD/Data/IHC Foundation Model/WSIs"
  output_directory: "C:/Users/Amaya/Documents/PhD/Data/IHC Foundation Model"
  embedding_weights: "C:/Users/Amaya/Documents/PhD/Data/WSI_foundation/"
  path_to_patches: "C:/Users/Amaya/Documents/PhD/Data/Test_Data_RA/extracted_patches_2/patches"
  unet_weights: "C:/Users/Amaya/Documents/PhD/Data/UNet_512_1.pth.tar"

# Dataset configurations
dataset:
  name: "RA"  # choices: ['RA', 'Sjogren']
  patch_size: 224
  overlap: 0
  coverage: 0.4
  slide_level: 1
  mask_level: 2
  patch_batch_size: 10
  train_fraction: 0.8
  val_fraction: 0.20
  stain_used: 'all' # RA choices: ['all', 'H&E', 'CD68', 'CD138', 'CD20']

# Parsing configurations
parsing:
  patient_ID: 'img.split("_")[0]' # "Patient123_stain.tiff" -> Patient123
  stain: 'img.split("_")[1]' # "Patient123_stain.tiff" -> stain
  stain_types: {'NA': 0, 'H&E': 1, 'CD68': 2, 'CD138': 3, 'CD20': 4} # RA stain types


# Label/split configurations
labels:
  label: 'label'
  label_dict: {'0': 'Pauci-Immune', '1': 'Lymphoid/Myeloid'}
  n_classes: 2
  patient_id: 'Patient_ID'

training:
  hidden_dim: 128
  learning_rate: 0.00001
  num_epochs: 5
  L2_norm: 0.00001
  batch_size: 1
  slide_batch: 1
  num_workers: 0
  scheduler: 1
  checkpoint: true
  seed: 42
  attention_heads: 1
  stratified_splits: 5

model:
  name: 'ABMIL'

# Execution flags
execution:
  preprocess: false
  segmentation: false
  embedding: false
  create_splits: false
  train: false
  test: false
  visualise: false

# Embedding models configs & paths
embedding_nets:
  # ImageNet pre-trained models
  vgg16:
    size: 4096
    weight_path: ""
    class: "VGG_embedding"
  resnet18:
    size: 512
    weight_path: ""
    class: "ResNet18_embedding"
  resnet50:
    size: 2048
    weight_path: ""
    class: "ResNet50_embedding"
  convnext:
    size: 1024
    weight_path: ""
    class: "ConvNext_embedding"
  ViT:
    size: 768
    weight_path: ""
    class: "ViT_embedding"
  # Foundation Models - Public data
  ssl_resnet18:
    size: 512
    weight_path: "Ciga/tenpercent_resnet18.pt"
    class: "ssl_resnet18_embedding"
  ssl_resnet50:
    size: 2048
    weight_path: ""
    class: "ssl_resnet50_embedding"
  CTransPath:
    size: 768
    weight_path: "CTransPath/ctranspath.pth"
    class: "CTransPath_embedding"
  Lunit:
    size: 384
    weight_path: ""
    class: "Lunit_embedding"
  GigaPath:
    size: 1536
    weight_path: ""
    class: "GigaPath_embedding"
  Phikon:
    size: 1024
    weight_path: ""
    class: "Phikon_embedding"
  # Foundation Models - Private data
  BiOptimus:
    size: 1536
    weight_path: ""
    class: "BiOptimus_embedding"
  UNI:
    size: 1024
    weight_path: ""
    class: "UNI_embedding"

# Mapping of class names to module paths
embedding_classes:
  VGG_embedding: "models.embedding_models.VGG_embedding"
  ResNet18_embedding: "models.embedding_models.ResNet18_embedding"
  ResNet50_embedding: "models.embedding_models.ResNet50_embedding"
  ConvNext_embedding: "models.embedding_models.ConvNext_embedding"
  ViT_embedding: "models.embedding_models.ViT_embedding"
  ssl_resnet18_embedding: "models.embedding_models.ssl_resnet18_embedding"
  ssl_resnet50_embedding: "models.embedding_models.ssl_resnet50_embedding"
  CTransPath_embedding: "models.embedding_models.CTransPath_embedding"
  Lunit_embedding: "models.embedding_models.Lunit_embedding"
  GigaPath_embedding: "models.embedding_models.GigaPath_embedding"
  Phikon_embedding: "models.embedding_models.Phikon_embedding"
  BiOptimus_embedding: "models.embedding_models.BiOptimus_embedding"
  UNI_embedding: "models.embedding_models.UNI_embedding"