{
  "docId": "019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "docSlug": "8757f1b44ef7f176",
  "documentTitle": "2024 Air Street Capital The State of AI Report 2024",
  "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": 83,
  "pageCount": 213,
  "prevPage": 82,
  "nextPage": 84,
  "slideType": "case_study",
  "function": "illustrate_case",
  "density": "dense",
  "nDataPoints": 0,
  "notes": "The slide highlights a specific research application of Stable Diffusion/U-Net for medical imaging.",
  "elementsJson": [
    "headline_text",
    "bullet_list",
    "paragraph",
    "photo"
  ],
  "metadataConfidence": 0.95,
  "imagePath": null,
  "slideHref": "/slides/019dd923-5e88-73ef-bd5d-0856e1444fb9/83",
  "deckHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9",
  "deckJsonHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9.json",
  "deckAnchorHref": "/decks/019dd923-5e88-73ef-bd5d-0856e1444fb9#slide-83",
  "components": [
    {
      "bbox": null,
      "kind": "callout",
      "text": "By jointly fine-tuning both the U-Net and the CLIP text encoder from Stable Diffusion a large dataset of real chest x-rays (CXR) and corresponding radiologist reports, it is possible to generate synthetic CXR scans with high fidelity and conceptual correctness as evaluated by board-certified radiologists.",
      "attrs": null,
      "subkind": null,
      "toolName": "Visual emphasis",
      "toolSlug": "visual-emphasis",
      "confidence": null,
      "componentId": "019dd952-47c3-7407-8d16-fac34bcd8f17",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.33,
        "x": 0.65,
        "y": 0.42
      },
      "kind": "image",
      "text": "Grid of 6 chest X-ray images with labels indicating pleural effusion status.",
      "attrs": null,
      "subkind": "illustration",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "afcfdea8-c0a7-4890-87cc-016396581174",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.45,
        "w": 0.6,
        "x": 0.02,
        "y": 0.42
      },
      "kind": "list",
      "text": "By jointly fine-tuning both the U-Net and the CLIP text encoder from Stable Diffusion a large dataset of real chest x-rays (CXR) and corresponding radiologist reports, it is possible to generate synthetic CXR scans with high fidelity and conceptual correctness as evaluated by board-certified radiologists.\nGenerated CXRs can be used for data augmentation and self-supervised learning.\nConsistent with other modalities, supervised classification performance drops slightly when training purely synthetic data.\nMoreover, generative models can improve fairness of medical classifiers by enriching training datasets with synthetic examples that fill out underrepresented data points.",
      "attrs": null,
      "subkind": "bullet",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5c770e26-60bb-435b-8ddb-4e2341fbadc1",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.18,
        "w": 0.6,
        "x": 0.02,
        "y": 0.22
      },
      "kind": "paragraph",
      "text": "High-quality medical imaging datasets are hard to come by or, even so, license to for research or commercial products. They are also not immune to distributional shifts. And yet, realistic image generators have flooded the internet in the last year. Could these be repurposed to generate realistic medical images that are useful for model training, despite the large visual and semantic differences between natural images and medical images?",
      "attrs": null,
      "subkind": "paragraph",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "74792e06-2951-40cb-ab58-237a57519383",
      "frameworkName": null,
      "frameworkSlug": null
    },
    {
      "bbox": {
        "h": 0.05,
        "w": 0.4,
        "x": 0.02,
        "y": 0.14
      },
      "kind": "title",
      "text": "Generating synthetic data in medicine",
      "attrs": null,
      "subkind": "headline",
      "toolName": null,
      "toolSlug": null,
      "confidence": null,
      "componentId": "5f63070c-3425-4761-856d-fec37ccbcbc9",
      "frameworkName": null,
      "frameworkSlug": null
    }
  ],
  "metrics": [],
  "tools": [
    {
      "name": "Visual Anchors",
      "slug": "visual-anchors",
      "agent": "Designer",
      "layer": "slide",
      "matchId": "ad858261-8981-4cd8-9db6-64e72efba659",
      "evidence": "Grid of 6 chest X-ray images with labels indicating pleural effusion status",
      "confidence": 0.6
    }
  ],
  "frameworks": [],
  "arcBeats": [
    {
      "to": 153,
      "from": 9,
      "beatId": "019dd95a-0682-776c-8e35-5f2398b8d1d0",
      "arcName": "The Triple Take",
      "arcSlug": "triple-take",
      "beatName": "The Facts (What)",
      "beatSlug": "triple-take-the-facts-what",
      "evidence": "Exec summary + Research + Industry sections inventory the year",
      "position": 1,
      "confidence": 78,
      "parentBeatName": "Setup",
      "parentBeatSlug": "setup"
    },
    {
      "to": 85,
      "from": 12,
      "beatId": "019dd95a-0682-776c-8e35-6d9d8ee06d0a",
      "arcName": "Voyage and Return",
      "arcSlug": "voyage-return",
      "beatName": "The Unknown",
      "beatSlug": "voyage-return-the-unknown",
      "evidence": "Research section explores frontier model uncharted territory",
      "position": 2,
      "confidence": 55,
      "parentBeatName": "Complication",
      "parentBeatSlug": "complication"
    }
  ],
  "loops": [
    {
      "to": 85,
      "from": 82,
      "name": "Pattern Hunter",
      "slug": "02-pattern-hunter",
      "bestFor": "Time-pressed audiences, building consensus, when data is strong",
      "matchId": "019dd95a-07fe-70ce-8d3e-4669f7c62bd4",
      "evidence": "Med-Gemini, synthetic CXR, FlowMind, balance-of-power slide closes section",
      "position": 10,
      "objective": "Round out research review with medicine + enterprise apps",
      "structure": "Evidence A -> Evidence B -> Evidence C -> Pattern/Conclusion",
      "confidence": 65,
      "description": "Group multiple pieces of evidence that together point to a pattern or conclusion"
    }
  ],
  "imagePathAlt": null,
  "thumbSrc": null,
  "thumbSrcAlt": null,
  "locked": true
}