Torsten Haferlach, MD, PhD
Just as artificial intelligence (AI) has seeped into our everyday existence through ChatGPT and AI-fueled smart speakers at home, the hematology field is also replete with opportunities that are only beginning to be explored — while also raising questions that are only beginning to be pondered.
A session at the 65th American Society of Hematology (ASH) Annual Meeting and Exposition will cover the possibilities of AI in detail, from helping with diagnoses to developing therapies and delving into related ethical questions.
Torsten Haferlach, MD, PhD, chair of the session and a co-founder of the Munich Leukemia Laboratory, which has embraced the use of AI, said he hopes the session will help people think carefully and constructively about the topic, rather than viewing it from one extreme or the other.
“There are barriers. There are people who are afraid or too enthusiastic,” he said. Instead “we just need to think about it.”
The 90-minute special interest session — AI in Hematology: Where Do You Stand in 2023? — will include four talks: “The Application of Artificial Intelligence in Medical Imaging: Lessons Learned from Radiology,” by Matthew Lungren, MD, PhD, of Microsoft Research; “Can AI Be Used in Drug Discovery?” by Weida Tong, PhD, of the U.S. Food & Drug Administration’s (FDA) National Center for Toxicological Research; “The Future is Now: AI in the Hematology Clinic,” by Dr. Haferlach; and “AI Ethics: More Complex Than What You Think,” by Yates Coley, PhD, of Kaiser Permanente Washington Health Research Institute.
The first talk will focus on how AI could be used to better and more efficiently employ imaging tools, such as those used to assess CT scans and other phenotypes.
“Could we improve the quality of our results or bring down the turnaround to the result and the diagnosis?” Dr. Haferlach said. “Those two things would at least be helpful.”
Another consideration, he said, is that AI could perform the tasks in medicine for which it can be difficult to find employees.
“In medicine, as in many other jobs we see right now, personnel are lacking,” he said. “And one question could be: Could we save personnel, or could we write the reports with fewer personnel?”
The FDA’s talk on drug discovery will look at how AI might expedite the introduction of new therapies, accelerating the pace of finding new drugs for new targets and fast-tracking the drug development process itself. The talk will assess whether AI could be a reliable tool to hurdle past the time-consuming evaluations that include cell models, mouse models, and phase I, II, and III clinical trials, a process that can take five to 10 years to move from an idea to an approved product.
One concept is the use of the so-called “digital twin,” which harnesses the power of huge datasets to create digital versions of real patients, matching their age, gender, and biological environment. These twins are then used to assess therapy options.
“Then you could quite easily decide about the drug development,” Dr. Haferlach said.
If the power of AI can be tapped in this way, it could streamline a product development process that often results in failed attempts, costing incredible amounts of money and time, he said.
“It’s such a long way, and it’s so expensive that many drugs stop on the way,” he said.
In his own talk, Dr. Haferlach will examine the power of AI to help make clinical management decisions by synthesizing vast amounts of information.
“There’s so much data available in PubMed, NCCN [National Comprehensive Cancer Center] guidelines, ICC [International Consensus Conference] and WHO [World Health Organization] classification, UpToDate, any online source … But we have to look it up and read it,” he said. However, AI can quickly process all these sources, he said, and “apply it to your question quite easily,” considering all of a patient’s health records.
“If you combine all these different sources, not in your brain, but by the support of an AI tool, that gives you an idea of what you should do,” he said. “I don’t say that’s always correct, but at least you have an idea. And this is improving every day.”
At Dr. Haferlach’s lab, data from 475,000 cases were used to power an AI tool to help diagnose patients. By analyzing what was essentially 475,000 questions, 475,000 answers, and 475,000 processes that led to those answers, the pilot tool successfully provided reports when presented with patient information, Dr. Haferlach said.
“You cannot discriminate between human-generated and AI-generated reports at the end,” he said. Processing all of the data for testing that is now part of the daily routine took three hours and about $3,000, he said.
Despite AI’s potential, ethical questions loom over its use in health care, he said, and this topic will be tackled in another talk.
“It’s not all good, it’s not all right,” Dr. Haferlach said. “So we have to be very careful. And the question is, do we have problems because of the diversity of the dataset? Do we only have Caucasian people in studies and never Black people? Do we have only rich people and not poor people?”
There is also the question of whether the AI process is too much of a mystery, he said.
“Many people still say AI is a black box, so you have to take it as it is,” he said but noted that it doesn’t have to be that way.
“What we apply here in the lab is called transparent AI, so the AI tells you, ‘Why did I come to this conclusion?’” he said.
He added that AI still falls under the supervision of physicians and that this responsibility should not change.
“There is no way to accept an AI result without control and final signature of the doctor or the nurse,” he said. “There’s always human control of any AI result in all the fields of medicine.”
The session will conclude with a half hour dedicated to questions and discussion from attendees, during which Dr. Haferlach suspects some of the most important ground will be covered.
“That will be very interactive,” he said. “This will be the most important 30 minutes.”
AI in Hematology: Where Do You Stand in 2023?
Monday, December 11, 10:30 a.m. - 12:00 p.m., San Diego Convention Center, Room 11, Upper Level