AI and Health Part One: DrGPT Will See You Now

Podcast: Babbage from The Economist
Host: Alok Jha
Release Date: May 22, 2024
Duration: Approximately 47 minutes

Introduction to AI in Healthcare

[01:07] Gerald Lipp, a radiologist overseeing breast cancer screening in Northeast Scotland, introduces the integration of artificial intelligence (AI) into mammogram analysis. Traditionally, two human readers evaluate each mammogram, with a third consulted in cases of disagreement. Gerald shares a compelling case where AI identified a subtle sign of cancer in an elderly patient, Barbara, which human readers had previously overlooked. This collaboration between AI and radiologists led to the early detection of a 5mm tumor, demonstrating AI's potential as an invaluable second pair of eyes in medical diagnostics.

Gerald Lipp:
“We wouldn’t have seen her again. And the AI did highlight a very subtle area in the right breast... Barbara was one of 11 women that we managed to pick up using the AI as an extra reader or a fail safe.”
[01:27]

The Expanding Role of AI in Diagnostics

Alok Jha elaborates on the burgeoning applications of AI in healthcare, from identifying fractures in X-rays to assisting in surgeries. However, Gerald emphasizes that while radiologists are proficient, AI provides an additional advantage by automating the review of numerous normal mammograms, thereby allowing more time for patient care.

Gerald Lipp:
“The AI gives us that little edge in screening... if I can automate and take away some of that 900 mammograms in 1000, that actually gives me more time to be patient.”
[03:21]

Generative AI and Its Potential in Medicine

[05:02] Natasha Loder, The Economist's health editor, discusses the surge of generative AI in medicine over the past year. She highlights its capabilities beyond diagnostics, including personalized health advice, patient triage, note-taking, meeting summaries, and drug discovery.

Natasha Loder:
“The possibilities lie beyond not just diagnostics, but also tools for giving personalized health advice... discovering new insights from medical data and do drug discovery.”
[05:20]

She identifies the primary challenge AI addresses: the global shortage of healthcare professionals amidst an aging population and rising healthcare costs.

Natasha Loder:
“There just aren't enough doctors or nurses... if we can allow these technologies to be harnessed, there's actually a lot of potential for improving productivity and meeting human need.”
[06:16]

Chiron Medical's AI Model: MIA

Peter Kachikamen, CEO of Chiron Medical, discusses MIA, an AI tool designed to enhance cancer detection in mammograms. Currently deployed in Hungary and the UK, MIA utilizes convolutional neural networks to analyze mammogram images, significantly increasing cancer detection rates by up to 13%.

Peter Kachikamen:
“AI can increase the cancer detection... with that, we can expect that we can actually halve the number of cancers missed in breast screening.”
[08:18]

He explains that MIA operates alongside radiologists, providing quick assessments and highlighting areas of concern without introducing bias, ensuring that final decisions remain under human oversight.

Peter Kachikamen:
“It's always a human radiologist who makes the final decision. They have the authority. The AI is a tool to flag when the AI is certain.”
[13:16]

Advancements in AI-Driven Radiology

Pranav Rajputkar, Assistant Professor at Harvard Medical School, delves into the evolution of AI in radiology. Initially focused on specific tasks like pneumonia detection in chest X-rays, his team now aims to develop generalist AI models capable of interpreting various imaging modalities and multiple diseases simultaneously.

Pranav Rajputkar:
“We're trying to build these AI models that can do a lot more than just one disease detection... a generalist medical AI perspective.”
[15:49]

He identifies three key advancements enabling this progress:

1. Self-Supervised Learning: Reduces the need for extensive labeled data by allowing AI to learn from unannotated datasets.
2. Multimodal Learning: Integrates data from different sources, such as imaging and clinical reports, to enhance understanding.
3. Generative AI: Produces comprehensive textual reports akin to those written by human radiologists.

Pranav Rajputkar:
“Generative AI allows us to make these much more interesting models that can start to look like what we expect from radiologists.”
[17:21]

Challenges and Ethical Considerations

As AI models grow more sophisticated, questions arise about their role in replacing human professionals. Pranav Rajputkar addresses the misconception that AI will replace radiologists, emphasizing the current shortage of medical professionals and the potential for AI to alleviate this gap.

Pranav Rajputkar:
“We just have a big supply shortage... AI models that can do these reads are going to be the solution we have on the horizon.”
[19:10]

He also touches upon the trust issues surrounding AI, comparing regulatory frameworks for AI to those for human doctors. However, Hugh Harvey, former Clinical Director at Chiron Medical, argues that AI should not be regulated like human doctors due to fundamental differences in accountability and operational dynamics.

Hugh Harvey:
“Large language models don't have moral accountability and they don't have... decades of experience like doctors do.”
[39:27]

Regulatory Hurdles for AI in Healthcare

Hugh Harvey discusses the stringent regulatory requirements AI models must meet to be approved as medical devices. He highlights the challenges generative AI poses, such as unpredictability and the inability to consistently demonstrate safety and effectiveness across infinite scenarios.

Hugh Harvey:
“We don't know how often it's going to get it wrong... Until we can measure that these things aren't safe to be on market as an intended medical device.”
[36:24]

He emphasizes the necessity for robust evidence and adherence to regulatory standards to ensure AI tools are both safe and effective for clinical use.

Patient Interactions with AI Chatbots

The podcast explores the practicality of using AI chatbots like ChatGPT for medical advice. Natasha Loder conducts a live test, revealing that while ChatGPT can provide general information, it lacks the specificity and reliability required for medical diagnoses. The conversation underscores the limitations of current AI models in handling sensitive health issues reliably.

Natasha Loder:
“We’re a long way technologically from being in a situation where we could recommend that anyone having these symptoms could just go and safely have a conversation with a chatbot.”
[28:36]

Alok Jha:
“How useful is that really in medical advice?”
[28:32]

The discussion points to the need for specialized, regulated medical chatbots that can offer accurate and consistent advice, potentially bridging the gap between general AI capabilities and clinical requirements.

Future Prospects and Conclusion

In concluding the episode, Natasha Loder envisions a future where AI serves as a first-line service for advice, triage, and diagnostics, especially in underserved areas. She anticipates AI becoming integrated with personal medical records, enabling highly personalized and efficient patient care.

Natasha Loder:
“AI has the potential to make medicine much more sophisticated, to reduce the amount of error, the misdiagnoses... It’s extraordinary to me how we could really start to meet the needs of so many more people with this technology.”
[41:27]

Alok Jha balances optimism with caution, acknowledging AI's transformative potential while recognizing the ethical, regulatory, and technical challenges that must be addressed to ensure its safe and effective implementation in healthcare.

Natasha Loder:
“In the short to medium term, it's less clear to me how rapidly these technologies will disseminate... I’m very excited. I can’t wait to see what’s coming down the road.”
[44:12]

The episode sets the stage for a forthcoming second part, which will delve deeper into AI’s impact on medical research and the future landscape of healthcare.

Key Takeaways

• AI as a Diagnostic Aid: AI models like Chiron Medical's MIA are proving effective in enhancing cancer detection rates, serving as valuable tools for radiologists.

• Generative AI's Expanding Role: Beyond diagnostics, generative AI holds promise for personalized health advice, efficient patient triage, and drug discovery.

• Regulatory Challenges: The integration of AI into healthcare faces significant regulatory hurdles, particularly for generative models that lack consistency and accountability.

• Ethical Considerations: Trust and reliability are paramount in medical AI applications, necessitating rigorous validation and oversight.

• Future Integration: AI has the potential to revolutionize healthcare by increasing access, reducing errors, and providing personalized care, especially in areas with limited medical resources.

Notable Quotes

• Gerald Lipp:
  “Barbara was one of 11 women that we managed to pick up using the AI as an extra reader or a fail safe.”
  [01:27]

• Natasha Loder:
  “The possibilities lie beyond not just diagnostics, but also tools for giving personalized health advice...”
  [05:20]

• Peter Kachikamen:
  “AI can increase the cancer detection... with that, we can expect that we can actually halve the number of cancers missed in breast screening.”
  [08:18]

• Pranav Rajputkar:
  “We're trying to build these AI models that can do a lot more than just one disease detection... a generalist medical AI perspective.”
  [15:49]

• Hugh Harvey:
  “Large language models don't have moral accountability and they don't have... decades of experience like doctors do.”
  [39:27]

• Natasha Loder:
  “AI has the potential to make medicine much more sophisticated, to reduce the amount of error, the misdiagnoses... It’s extraordinary to me how we could really start to meet the needs of so many more people with this technology.”
  [41:27]

Conclusion

The episode "AI and Health Part One: DrGPT Will See You Now" provides a comprehensive exploration of how AI is transforming healthcare. From enhancing diagnostic accuracy to addressing the global shortage of medical professionals, AI's potential is vast. However, realizing this potential requires overcoming significant regulatory, ethical, and technical challenges. As AI continues to evolve, its integration into healthcare promises to revolutionize patient care, making it more accessible, efficient, and personalized.

Produced by Jason Hoskin and Lawrence Knight. Mixing and sound design by Nico Rofast. Executive Producer: Hannah Mourinho.