Physician CEO discusses benefits, challenges to AI in medicine at FTC hearing

Michael Abramoff, MD, PhD
Michael Abramoff

“Carried out in a responsible manner, autonomous artificial intelligence has the promise of lowering cost and increasing accessibility and quality in health care,” IDx founder and CEO, Michael Abramoff, MD, PhD, said at an FTC hearing on ethical and consumer protection issues related to new AI applications.

Abramoff was invited to lend his expertise as a practicing physician and founder of IDx, which received FDA clearance for an autonomous AI diagnostic system.

He spoke during a section of the hearing entitled, “Understanding Algorithms, Artificial Intelligence, and Predictive Analytics Through Real-World Applications.”

When asked about what circumstances permit the use of AI technologies over traditional algorithms and vice versa, Abramoff said AI is highly dependent on the intended use and deployment conditions.

“For specific types of medical applications, AI technology achieves higher performance on cognitively complex tasks than most humans, and there are many algorithmic approaches, including deep learning, that AI can be built with,” according Abramoff in the hearing transcript. “AI technologies can improve accuracy and performance over traditional hard-coded algorithms and perform better on highly complex optimization problems.”

He added that when selecting AI technology, “one must concurrently assess practical, policy and ethical issues in addition to technical and use case considerations.”

Abramoff was asked what factors, resources or polices have facilitated the development or advancement of these technologies. He said the biggest challenge continues to be the scarcity of good quality data.

“Thus, for a long time, accuracy could not reach an acceptable level for patient safety. Not because we did not have the algorithms, but primarily because the input data was so noisy, as it was primarily based on manual entry of doctors’ individual impressions,” he said.

He cited the development of low-cost sensors, which have increased the quality input vectors dramatically and “allowing for accuracy as we see it today – and then we realized that the reference standard, usually created by clinicians, was a problem,” he said.

Abramoff added that his own research was supported for years by NIH, and he worked closely with the FDA, “to advance the safety and validation of autonomous AI in medicine.”

As far as policies or other initiatives that have impeded the development of AI, he listed four factors.

First, while the FDA approval of IDx-DR took years of discussion, the template is now available for any company to obtain authorization for autonomous AI, he said.

Second, the lack of reimbursement models can be a challenge. However, IDx-DR has a pathway to payment, and they are currently seeking further coverage clarification from CMS and other commercial insurers, Abramoff said.

Third, he cited interoperability of autonomous AI with electronic health records as crucial for increasing clinic efficiency. Abramoff encouraged work by the National Coordinator for Health Information Technology to improve interoperability.

Finally, initial product reviews for new AI technologies should still require a preregistered clinical trial “with a ‘locked’ down, deterministic AI that accounts for human factors and system-level validation, such as use training, product labelling and system workflow,” he added.

“Consumers, patients, and physicians cannot make a determination of AI quality without being given visibility to the underlying architecture and rigorous testing,” Abramoff said later in hearing.

He was asked about the concern of “human control” of artificial intelligence and whether he believes AI will remain in human control.

“In medicine, AI is part of a larger health care ‘system’ rather than a product, so the clinical validation should ensure that the level of ‘human control’ either maintains or improves safety and efficacy after introducing the AI system,” he said.

Abramoff has previously provided his expertise on AI in health care to the U.S. Congress and the White House Office of Science and Technology, according to a press release from IDx.

Reference:

FTC Hearing #7 on algorithms, artificial intelligence, and predictive analytics. November 13, 2018. Accessed December 29, 2018.

Disclosure: Abramoff is founder and CEO of IDx.

 

Michael Abramoff, MD, PhD
Michael Abramoff

“Carried out in a responsible manner, autonomous artificial intelligence has the promise of lowering cost and increasing accessibility and quality in health care,” IDx founder and CEO, Michael Abramoff, MD, PhD, said at an FTC hearing on ethical and consumer protection issues related to new AI applications.

Abramoff was invited to lend his expertise as a practicing physician and founder of IDx, which received FDA clearance for an autonomous AI diagnostic system.

He spoke during a section of the hearing entitled, “Understanding Algorithms, Artificial Intelligence, and Predictive Analytics Through Real-World Applications.”

When asked about what circumstances permit the use of AI technologies over traditional algorithms and vice versa, Abramoff said AI is highly dependent on the intended use and deployment conditions.

“For specific types of medical applications, AI technology achieves higher performance on cognitively complex tasks than most humans, and there are many algorithmic approaches, including deep learning, that AI can be built with,” according Abramoff in the hearing transcript. “AI technologies can improve accuracy and performance over traditional hard-coded algorithms and perform better on highly complex optimization problems.”

He added that when selecting AI technology, “one must concurrently assess practical, policy and ethical issues in addition to technical and use case considerations.”

Abramoff was asked what factors, resources or polices have facilitated the development or advancement of these technologies. He said the biggest challenge continues to be the scarcity of good quality data.

“Thus, for a long time, accuracy could not reach an acceptable level for patient safety. Not because we did not have the algorithms, but primarily because the input data was so noisy, as it was primarily based on manual entry of doctors’ individual impressions,” he said.

He cited the development of low-cost sensors, which have increased the quality input vectors dramatically and “allowing for accuracy as we see it today – and then we realized that the reference standard, usually created by clinicians, was a problem,” he said.

Abramoff added that his own research was supported for years by NIH, and he worked closely with the FDA, “to advance the safety and validation of autonomous AI in medicine.”

As far as policies or other initiatives that have impeded the development of AI, he listed four factors.

First, while the FDA approval of IDx-DR took years of discussion, the template is now available for any company to obtain authorization for autonomous AI, he said.

PAGE BREAK

Second, the lack of reimbursement models can be a challenge. However, IDx-DR has a pathway to payment, and they are currently seeking further coverage clarification from CMS and other commercial insurers, Abramoff said.

Third, he cited interoperability of autonomous AI with electronic health records as crucial for increasing clinic efficiency. Abramoff encouraged work by the National Coordinator for Health Information Technology to improve interoperability.

Finally, initial product reviews for new AI technologies should still require a preregistered clinical trial “with a ‘locked’ down, deterministic AI that accounts for human factors and system-level validation, such as use training, product labelling and system workflow,” he added.

“Consumers, patients, and physicians cannot make a determination of AI quality without being given visibility to the underlying architecture and rigorous testing,” Abramoff said later in hearing.

He was asked about the concern of “human control” of artificial intelligence and whether he believes AI will remain in human control.

“In medicine, AI is part of a larger health care ‘system’ rather than a product, so the clinical validation should ensure that the level of ‘human control’ either maintains or improves safety and efficacy after introducing the AI system,” he said.

Abramoff has previously provided his expertise on AI in health care to the U.S. Congress and the White House Office of Science and Technology, according to a press release from IDx.

Reference:

FTC Hearing #7 on algorithms, artificial intelligence, and predictive analytics. November 13, 2018. Accessed December 29, 2018.

Disclosure: Abramoff is founder and CEO of IDx.