서브메뉴
검색
Development, Evaluation, and Deployment of Medical AI
Development, Evaluation, and Deployment of Medical AI
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105618
- ISBN
- 9798265428547
- DDC
- 300
- 저자명
- Wu, Eric.
- 서명/저자
- Development, Evaluation, and Deployment of Medical AI
- 발행사항
- [Sl] : Stanford University, 2025
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2025
- 형태사항
- 190 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Zou, James.
- 학위논문주기
- Thesis (Ph.D.)--Stanford University, 2025.
- 초록/해제
- 요약Artificial intelligence (AI) holds incredible promise to transform clinical care, yet its safe and effective deployment on patients depends on robust evaluation and domain-specific advancements. This dissertation advances those goals through eight peer-reviewed studies that span the full lifecycle of medical AI. First, we present a trio of works revealing limitations in the regulatory and deployment landscape of medical AI devices. Our systematic review of FDA-cleared AI devices show that 1) approvals have been based primarily single-site, retrospective studies; 2) post-market monitoring and model updating are rare; and 3) only a small subset of regulatory-approved algorithms are ever utilized in clinical care. Second, we propose three benchmarks that highlight current shortcomings in medical large language models (LLMs): 1) models often adopt erroneous retrieved text over correct prior knowledge, 2) LLMs often hallucinate supporting citations, and 3) are unable to reliably learn new and updated medical knowledge via fine-tuning. Third, moving beyond evaluation to spatial biology, we introduce 7-UP and ROSIE, complementary works that leverage generative AI to impute high-dimensional molecular information from routine clinical assays. By training deep learning models on millions of imaged cells, we can perform in silico staining to obtain dozens of spatially resolved protein markers from only H\\&E or low-plex immunofluorescence panels. Collectively, these contributions reveal shortcomings in current medical AI regulation, introduce robust benchmarks and evaluations of clinical LLM deployment, and present generative methods that widen access to high-dimensional tissue profiling. The work provides evidence-based recommendations for regulators, developers, and clinicians, and delivers open-source tools and datasets to accelerate safe and effective adoption of AI in medicine.
- 일반주제명
- Patients
- 일반주제명
- Medical equipment
- 일반주제명
- Citations
- 일반주제명
- Clinical medicine
- 일반주제명
- Large language models
- 일반주제명
- Chatbots
- 일반주제명
- Medicine
- 기타저자
- Stanford University.
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2025 us c eng d■001000017360781
■00520260202105618
■006m o d
■007cr#unu||||||||
■020 ▼a9798265428547
■035 ▼a(MiAaPQ)AAI32316484
■035 ▼a(MiAaPQ)Stanfordkd989mp6611
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a300
■1001 ▼aWu, Eric.
■24510▼aDevelopment, Evaluation, and Deployment of Medical AI
■260 ▼a[Sl]▼bStanford University▼c2025
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2025
■300 ▼a190 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Zou, James.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2025.
■520 ▼aArtificial intelligence (AI) holds incredible promise to transform clinical care, yet its safe and effective deployment on patients depends on robust evaluation and domain-specific advancements. This dissertation advances those goals through eight peer-reviewed studies that span the full lifecycle of medical AI. First, we present a trio of works revealing limitations in the regulatory and deployment landscape of medical AI devices. Our systematic review of FDA-cleared AI devices show that 1) approvals have been based primarily single-site, retrospective studies; 2) post-market monitoring and model updating are rare; and 3) only a small subset of regulatory-approved algorithms are ever utilized in clinical care. Second, we propose three benchmarks that highlight current shortcomings in medical large language models (LLMs): 1) models often adopt erroneous retrieved text over correct prior knowledge, 2) LLMs often hallucinate supporting citations, and 3) are unable to reliably learn new and updated medical knowledge via fine-tuning. Third, moving beyond evaluation to spatial biology, we introduce 7-UP and ROSIE, complementary works that leverage generative AI to impute high-dimensional molecular information from routine clinical assays. By training deep learning models on millions of imaged cells, we can perform in silico staining to obtain dozens of spatially resolved protein markers from only H\\&E or low-plex immunofluorescence panels. Collectively, these contributions reveal shortcomings in current medical AI regulation, introduce robust benchmarks and evaluations of clinical LLM deployment, and present generative methods that widen access to high-dimensional tissue profiling. The work provides evidence-based recommendations for regulators, developers, and clinicians, and delivers open-source tools and datasets to accelerate safe and effective adoption of AI in medicine.
■590 ▼aSchool code: 0212.
■650 4▼aPatients
■650 4▼aMedical equipment
■650 4▼aCitations
■650 4▼aClinical medicine
■650 4▼aLarge language models
■650 4▼aChatbots
■650 4▼aMedicine
■690 ▼a0800
■690 ▼a0564
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2025
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360781▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


