Teaching Machines How to Smell Disease — Lessons Learned
summary
Claim: breath can screen for lung cancer but only with tight engineering and clinical care. Prototype uses tunable diode laser spectroscopy plus spirometry, yields rapid per‑test reads and low marginal cost; another arm runs standard lab methods. A 70-patient study produces useful signals but also forces discarding candidate biomarkers. Key failures trace to sampling: humidity and condensation (molecules condense below ~34°C), confounders like smoking, and regulatory and explainability gaps. Practical path: niche, monetizable use case first.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 13:04Breath analysis complexity
- 20:45Magical thinking in AI
- 27:27The need for explainability over explainability
- 33:10Risk stacking in diagnostics
- 36:25Smell-based diagnostics state
- 44:54Q&A
related talks
19mAI, Technology & Health: What's Smell Got to Do With It?
30mAIMED: AI for Prostate Cancer Detection and Immunotherapy Response
39mBridging health inequalities through machine learning
34mEnabling Early Stage Breast Cancer Detection Using AI
26mToward Affordable Preventive Care
01h 01mVocal biomarkers for disease detection
06mListen2Cough: Leveraging End-to-End Deep Learning Cough Detection Model to Enhance Lung Health ...
24mMachine Learning & Medical Imaging: The Future of Early Cancer Detection