{
  "field_id": "F044",
  "field": "computational pathology",
  "directions": [
    {
      "direction_id": "G0",
      "name": "Transfer and resilience of pathology image models",
      "scope": "Study how pathology-image features can be reused across multiple downstream analyses and how model performance can be maintained or improved when stains, tissue preparation, imaging, or data sources change. This includes transferable encoders, self-supervised and pathology-specific feature learning, domain adaptation, domain-shift evaluation, and augmentation designed for distribution shifts."
    },
    {
      "direction_id": "G1",
      "name": "Clinical and biological inference from histology",
      "scope": "Infer case status from pathology images and derived tissue measurements, including morphological diagnosis and grading; molecular status, continuous biomarkers, and gene expression; and survival, progression, malignant transformation, or other future risks. Molecular predictions may also use clinical variables."
    },
    {
      "direction_id": "G2",
      "name": "Spatial identification of tissue and microscopic structures",
      "scope": "Locate, segment, identify, or quantify tissue compartments and lesions, as well as individual cells, nuclei, mitoses, and localized microscopic biomarkers in pathology images, to obtain spatial or quantitative results for the structures themselves."
    },
    {
      "direction_id": "G3",
      "name": "Histological image and morphology synthesis",
      "scope": "Synthesize histologically informative images or morphology: either generate faithful target stains and imaging readouts from existing stained or label-free tissue images, or generate new tissue, cell, or nuclear structures and samples, potentially conditioned on layouts, annotations, labels, or outcomes."
    },
    {
      "direction_id": "G4",
      "name": "Appraisal and explanation of pathology model decisions",
      "scope": "Examine pathology-model performance and the basis for model decisions through systematic comparisons of model choices and evaluation methods, analyses of agreement between models and pathologists, and studies that explain predictions in terms of recognizable regions, cellular features, concepts, or explicit reasoning."
    },
    {
      "direction_id": "G5",
      "name": "Correspondence across tissue slides",
      "scope": "Establish matched spatial correspondence between tissue images from different sections or stains, including whole-slide geometric registration and matched multi-stain resources whose principal purpose is aligned correspondence."
    }
  ]
}