EdikK/hermes-literature

Literature Index — AI in Healthcare / Clinical Decision Support

This directory contains source PDFs and extracted texts for preparing scientific articles on AI in healthcare, clinical decision support systems (CDSS), machine learning, and related topics.

Source PDFs (source_pdfs/)

#	File	Size	Language	Topic
1	EAA_Module1.pdf	~333 KB	EN	Stanford: AI in Healthcare (Module 1)
2	EAA_Module2.pdf	~312 KB	EN	Stanford: Evaluations of AI in Healthcare (Module 2)
3	EAA_Module3.pdf	~277 KB	EN	Stanford: AI Deployment (Module 3)
4	EAA_Module4.pdf	~188 KB	EN	Stanford: Bias and Fairness (Module 4)
5	EAA_Module5.pdf	~427 KB	EN	Stanford: Regulatory Environment (Module 5)
6	Fundamentals of ML Study Guide.pdf	~5.9 MB	EN	Stanford: Fundamentals of AI/ML for Healthcare (full)
7	Fundamentals of ML Study Guide - M4.txt	~1.4 MB	EN	Module 4: Evaluation and Metrics
8	Fundamentals of ML Study Guide - M5.txt	~484 KB	EN	Module 5: Strategies and Challenges
9	Fundamentals of ML Study Guide - M6.txt	~455 KB	EN	Module 6: Best Practices, Teams, Launch
10	Intro to Clinical Data Study Guide - M6.txt	~213 KB	EN	Module 6: Electronic Phenotyping
11	Intro to Clinical Data Study Guide - M7.txt	~251 KB	EN	Module 7: Clinical Data Ethics
12	nihms-1033618.pdf	~160 KB	EN	Rajkomar et al. — Ensuring Fairness in ML to Advance Health Equity (Ann Intern Med, 2018)
13	tacca122000048.pdf	~176 KB	EN	Densen P. — Challenges and Opportunities Facing Medical Education (Trans Am Clin Climatol Assoc, 2011)
14	bmj00449-0050.pdf	~1.3 MB	EN	BMJ: Gillon "four principles plus scope" + Altman & Bland ROC plots
15	The_Elements_of_Statistical_Learning_Dat.pdf	~12.2 MB	EN	Hastie, Tibshirani, Friedman — The Elements of Statistical Learning (2nd ed.)
16	«Глубокое обучение», Ян Гудфеллоу.pdf	~14.9 MB	RU	Goodfellow, Bengio, Courville — Deep Learning (Russian edition, DMK Press)

Extracted Texts (extracted/)

Same file names with .txt or .md extension, extracted via pdftotext or provided as text by user.

Newly added (user-sent text)

• bmj00449-0050.md — Gillon medical ethics (four principles + scope) + Altman & Bland ROC curves
• «Глубокое обучение»,_Ян_Гудфеллоу.md — Goodfellow DL book front matter (RU)

Key Concepts for CDSS / AI in Healthcare Articles

• Outcome-Action Pairing (OAP): AI model output must be paired with a concrete clinical action; without action there is no clinical utility (Stanford EAA M2–M3).
• Clinical utility beyond accuracy: decision curve analysis, net benefit analysis, number needed to screen/treat/harm (Stanford EAA M2, Fundamentals of ML M4).
• SaMD risk categories (IMDRF/FDA): Categories I–IV; III–IV require pre-market approval (Stanford EAA M5).
• EU AI Act 2024/1689: medical AI = high-risk system; conformity assessment, transparency, human oversight.
• FDA CDS Guidance (Jan 2026): boundary between supportive tools and SaMD.
• Bias types: historical, representation, measurement, aggregation, evaluation, deployment bias; label bias, minority bias, privilege bias, automation bias, dismissal bias (Rajkomar et al. 2018; Stanford EAA M4).
• Distributive justice: equal outcomes, equal performance, equalized odds, predictive parity (Rajkomar et al. 2018).
• Medical knowledge doubling: 1950 — 50 years; 1980 — 7 years; 2010 — 3.5 years; 2020 — 73 days (Densen 2011).
• MINIMAR: minimum information for medical AI reporting.
• Medical ethics — four principles + scope (Gillon, BMJ 1994): autonomy, beneficence, non-maleficence, justice; scope of application determines to whom/what obligations apply. Underpins CDSS human-in-the-loop and responsibility model.
• ROC curves (Altman & Bland, BMJ 1994): sensitivity vs 1-specificity; AUC = probability that a random diseased person scores higher than a random non-diseased person. Supports "clinical utility beyond accuracy" section.

Usage

Use these sources when preparing articles on:

• Clinical Decision Support Systems (CDSS)
• AI in healthcare evaluation and deployment
• Bias and fairness in medical AI
• Regulatory frameworks (EU AI Act, FDA SaMD)
• Machine learning fundamentals for healthcare

Knowledge Graph (knowledge_graph/)

Separate graph for literature navigation and article preparation:

• literature_kg.json — 28 nodes, 33 edges, 7 clusters (regulatory, evaluation, bias/fairness, deployment, foundations, databases, architecture).
• literature_kg.mmd — Mermaid diagram for visual rendering.

Core node: CDSS Architecture (3-layer + Human-in-the-loop) connects to all regulatory, evaluation, bias, deployment, foundation, and database nodes.

Clusters:

• Regulatory: EU AI Act 2024/1689, FDA CDS Guidance, FDA SaMD, Stanford EAA M5
• Evaluation: OAP, GRADE, MINIMAR, Stanford EAA M2, FundML M4
• Bias/Fairness: Rajkomar 2018, Stanford EAA M4, ClinData M7 Ethics
• Deployment: Stanford EAA M3, FHIR, SMART, ClinData M6 Phenotyping
• Foundations: Elements of Statistical Learning, Глубокое обучение (RU), FundML M5
• Databases: PubMed, Cochrane, ClinicalTrials.gov, ChEMBL, DrugBank
• Architecture: PheKB, OHDSI/Aphrodite

Author

Eduard Koshilko — Ukrainian-trained MD, Approbation/FSP in progress (DE), not licensed in DE/EU, no clinical practice. Haftungsausschluss applies.