Panel
Transforming Healthcare: Unleashing GenAI's Potential Across Medical Frontiers | RAISE Summit 2024
- Panel Focus & Context: The session "Transforming Healthcare: Unleashing Gen AI's Potential" addressed the shift from doomy predictions to practical complementarity between AI and medical professionals, featuring founders from Hope Valley AI, Glimmer, DeepSyn, and OKI.
- Precision Medicine & Drug Discovery (Alban de la Sabliere, OKI):
- AI applications in pharma rely on "knowledge graphs on steroids" to manage complex biological data and mathematical power to accelerate 3D protein design.
- AlphaFold serves as a key example, reducing 3D protein design time from years to minutes based on RNA sequences.
- Gen AI addresses the "reverse translational gap" by modeling human biological complexity, a hurdle where mouse cancer cures (developed 30 years ago) failed to translate to humans.
- OKI has developed a diagnostic test predicting mutations from pathology slides with accuracy equal to or better than traditional biological tests, currently approved in Europe.
- Radiology & Clinical Efficiency (Christian Allouche, Glimmer):
- Glimmer's AI "co-pilot" for radiologists targets two primary value propositions: increasing reading speed to address global shortages and reducing false negative rates.
- Deployment statistics include 1,200 clinical sites across 34 countries (primarily US and Europe), processing over 20 million patient exams annually.
- Clinical evidence indicates a 30% average reduction in radiologist errors, with specific error reductions up to 50% for lung nodules on chest X-rays.
- Meta-analysis of production data estimates the software prevents approximately 500,000 missed fractures per year.
- Strategic roadmap involves extending coverage to mammography and CT scans within the next few years.
- Accessibility & Mobile Imaging (Clément Stippel, DeepSyn):
- Current global access to advanced medical imaging is limited, with seven out of 10 people lacking access to devices like MRI.
- DeepSyn's solution utilizes AI to create MRI systems that are 10 times more affordable, fully mobile, and open-source.
- The technology aims to serve the five billion people currently without access, impacting both developing nations and developed markets with long waiting times.
- Applications include enabling preventative healthcare by democratizing access to imaging in remote or underserved areas.
- Investment & Market Analysis (Charles Roberts, ARK Invest):
- Investors are actively filtering "real" AI solutions from "smoke and mirrors," distinguishing between companies with genuine research/talent versus those piggybacking on public APIs like GPT-4.
- Key investment thesis focuses on the "wheat from chaff" between point solutions (e.g., reading colonoscopy reports) and potential end-to-end transformative capabilities.
- Freenome, an ARC Venture portfolio company, utilizes Gen AI for natural language processing of clinical reports; its pivotal study showed ~60% detection for Stage 1 colorectal cancer, a life-saving metric for a disease killing nearly one million globally.
- The primary long-term opportunity identified is "causal discovery" in biology, using GPT-like models to understand high-dimensional, non-linear biological relationships rather than just correlation.
- Trust, Ethics, & Regulation:
- A 2023 study indicates 60% of Americans feel uncomfortable when AI providers directly use AI in diagnostics or treatment without human intervention.
- Human Oversight: Consensus dictates AI must act as a "co-pilot" rather than a replacement, with final diagnostic decisions resting with healthcare providers to satisfy trust requirements.
- Regulatory Landscape:
- Medical device regulations (MDR) and the new EU AI Act require rigorous clinical data and audited compliance, with European notified bodies creating bottlenecks (9-month audit wait + 6-8 month certification).
- The US FDA is described as faster but setting higher safety bars.
- Data & GDPR Conflict: The EU AI Act requires transparency on training data demographics (e.g., ethnicity), which conflicts with GDPR restrictions on collecting personal data, hindering the proof of model generalization.
- Explainability: Solutions like DeepSyn's "CADIC" software highlight detected findings (e.g., "lung nodule is there") rather than making autonomous diagnoses, allowing experts to validate or challenge AI findings.
- Diversity & Bias:
- Industry leaders acknowledge the historical male dominance in AI and biotech (e.g., ~93% male enrollment at Imperial College historically) and note progress is "too slow."
- Impact on Outcomes: Diversity in development teams is critical for reducing algorithmic bias, particularly in conditions affecting specific demographics (e.g., breast cancer).
- Leadership Action: Companies like Relation Therapeutics and Freenome have actively sought female leadership (e.g., female CSOs) to improve decision-making and reduce "groupthink."
- Diversity of Thought: Roberts notes that LLM performance improves with diverse prompting contributors (including liberal arts backgrounds), suggesting non-technical diversity is essential for better AI results.
- Cultural Retention: Alban de la Sabliere emphasizes that diversity requires a supportive culture to prevent "optical diversity" where individuals feel pressured to conform to a male-dominated mindset.