{
  "docId": "019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "docSlug": "bf350dd574c19997",
  "documentTitle": "2021 Air Street Capital The State of AI Report 2021",
  "authorId": "AirStreetCapital",
  "authorName": "Air Street Capital",
  "documentKindSlug": "consulting-deck",
  "documentKindLabel": "Consulting deck",
  "sourceTypeSlug": "vc_research",
  "sourceTypeLabel": "VC research",
  "presentationDate": null,
  "orientation": "landscape",
  "aspectRatio": 1.777,
  "pageNumber": 55,
  "pageCount": 188,
  "prevPage": 54,
  "nextPage": 56,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 24,
  "notes": "The slide highlights a specific research finding regarding algorithmic bias in medical imaging.",
  "elementsJson": [
    "headline_text",
    "paragraph",
    "bullet_list",
    "data_table"
  ],
  "metadataConfidence": 1,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5c-fd5012384ae3/55",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5c-fd5012384ae3#slide-55",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "This implies that medical AI systems can potentially cause discriminatory harm and reproduce or exacerbate the racial disparities that already exist in medical practice.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d2b-67e83445219f",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.58,
        "x": 0.03,
        "y": 0.41
      },
      "kind": "list",
      "text": "A multi-site study using public and private chest X-ray, chest CT, digital radiography, breast mammogram, and spine X-ray image data were used to built AI systems for race detection.\nTrained models displayed >0.8 and often >0.9 ROC-AUC scores on the task of race prediction across imaging modalities, suggesting very high performance on this task.\nWorryingly, this detection is not due to trivial proxies, such as body habitus, age, or other potential imaging confounders.\nLearned features appear to involve all regions of the image and frequency spectrum, which complicates mitigation efforts.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "b56dee35-de9f-4600-88ac-4d52f693453e",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": null,
      "kind": "metric",
      "text": "ROC-AUC: 0.9",
      "attrs": null,
      "subkind": "primary",
      "toolName": "Quantification",
      "toolSlug": "quantification",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d2b-6a2928f85b75",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.15,
        "w": 0.94,
        "x": 0.03,
        "y": 0.21
      },
      "kind": "paragraph",
      "text": "There is a conundrum in medical imaging AI: While computer vision models trained on a patient's medical imaging data of various modalities can accurately and trivially predict their race, clinicians attempting to do the same cannot. This implies that medical AI systems can potentially cause discriminatory harm and reproduce or exacerbate the racial disparities that already exist in medical practice.",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "13c48955-c93c-4795-8c0e-3b8b6a7a8dc0",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.35,
        "x": 0.627,
        "y": 0.418
      },
      "kind": "table",
      "text": "Experiments, Dataset, ROC-AUC (Black), Location of results. A, Race detection in radiology imaging. A1, CXR - internal validation, MXR(Resnet34/Densenet121) CXP(Resnet34) EMX(Resnet34/Densenet121/EfficientNet-B0), 0.97/0.95 0.98 0.98/0.99/0.99, Main text. CXR - external validation, MXR to CXP/MXR to EMX CXP to EMX/CXP to MXR EMX to MXR/EMX to CXP, 0.97/0.97 0.97/0.96 0.98/0.98, Main text. CXR - comparison of models, MXR/CXP/EMX, Multiple results, Supplement. A2, CT chest - internal validation, NLST(slice/study), 0.92/0.96, Main text. CT chest - external validation, NLST to EM-CT(slice/study) NLST to RSPECT(slice/study), 0.80/0.87 0.83/0.90, Main text. Limb x-ray - internal validation, DHA, 0.91, Main text. Mammography, EM-Mammo(image/study), 0.82/0.84, Main text. Cervical spine x-ray, EM-CS, 0.92, Main text. A3, CXR - models trained for, MXR - pathology detection task, 0.86, Main text.",
      "attrs": null,
      "subkind": "data",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "bb2aca70-c808-4b28-b3de-c8127b0b79df",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.04,
        "w": 0.94,
        "x": 0.03,
        "y": 0.14
      },
      "kind": "title",
      "text": "Medical AI racism: models reliably identify the self-reported racial identity of patients",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "c8548020-e1dc-4643-b797-b3bdc864c14f",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Data-Ink Ratio",
      "slug": "data-ink-ratio",
      "agent": "Designer",
      "layer": "slide",
      "matchId": "e67bac14-a97c-4870-817d-9b8938af8ed9",
      "evidence": "Table/data: Experiments, Dataset, ROC-AUC (Black), Location of results",
      "confidence": 0.6
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 152,
      "from": 5,
      "beatId": "019dd95a-0682-776c-8e35-0affbadec38e",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Definitions p.5-6, Exec Summary p.7, then Sections 1-3 catalog evidence.",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 152,
      "from": 10,
      "beatId": "019dd95a-0682-776c-8e35-1b3ee3a1c012",
      "arcName": "The Mountain",
      "arcSlug": "mountain",
      "beatName": "Rising Action",
      "beatSlug": "mountain-rising-action",
      "evidence": "Three sections accumulating evidence of accelerating progress.",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Development",
      "parentBeatSlug": "development"
    }
  ],
  "loops": [
    {
      "to": 55,
      "from": 52,
      "name": "Iceberg",
      "slug": "10-iceberg",
      "bestFor": "Consulting, complex problem solving, organizational change",
      "matchId": "019dd95a-07fe-70ce-8d3d-5a2841b69367",
      "evidence": "Data deserts -> measuring bias -> 94% systems worse than radiologist -> racial identification — symptom to system to root.",
      "position": 7,
      "objective": "Surface bias problems in medical AI",
      "structure": "The Symptom (Visible) -> The System (Hidden) -> The Root Cause",
      "confidence": 78,
      "description": "Reveal that the visible problem is merely a symptom of a deeper root cause"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}