麻豆果冻传媒影音

VIENNA -- Researchers trained an artificial intelligence (AI) model to determine if a patient with hepatocellular carcinoma (HCC) will respond to standard treatment with atezolizumab (Tecentriq) plus bevacizumab (Avastin).

By applying various algorithms to multiple datasets, progression-free survival (PFS) was improved among patients with high atezolizumab-bevacizumab receptor signature (ABRS) scores compared with patients with low scores, reported Qinghe Zeng, a PhD student at Laboratoire d'Informatique Paris Descartes at Université Paris Cité, during her late-breaker oral presentation at the European Association for the Study of the Liver annual meeting.

At 1 year, those with high scores had a PFS rate of 50% compared with a rate of 25% for those with low scores (P=0.0014).

There was no statistically significant difference in overall survival.

"This is a proof-of concept study and requires prospective evaluation," Zeng said. "Our deep learning model is able to predict an ABRS expression in resection and biopsy slides. We have proven that it can be predictive of progression-free survival in treated patients."

HCC is the most common primary liver cancer. While atezolizumab, an immune checkpoint inhibitor, combined with bevacizumab, an anti-angiogenic agent, is one of the standard-of-care treatments for advanced disease, only a subset of patients respond, Zeng said.

Julien Calderaro, MD, PhD, of Henri Mondor Hospital in Paris and the principal researcher on the project, told MedPage Today at a press conference that "this project is not ready for clinical use at this time, and it is not likely that we will see artificial intelligence or deep learning in the liver disease field for a few years, probably at least 5 years."

He suggested, however, that with use of the ABRS score, physicians may eventually be able to more precisely select which patients are likely to benefit from atezolizumab and bevacizumab, and, "more importantly, which patients would do better seeking alternative treatments."

The researchers used histological slides and multiple datasets -- a discovery series based on 336 resections from multiple institutions, an external validation dataset based on 225 resection samples from Henri Mondor Hospital, an external validation dataset of 157 biopsy samples from two French hospitals, an external validation dataset of 122 samples from atezolizumab- and bevacizumab-treated patients, and an external validation set of four in situ spacial transcriptomics from resections done at Henri Mondor Hospital -- to produce the gene signature biomarker.

Press conference moderator Aleksander Krag, MD, PhD, of the University of Southern Denmark in Odense, told MedPage Today that "this obviously is early days for AI in liver disease, but what I think it tells us is that we've opened the door to a new world, where we can see that some of this very powerful technology can improve patient care," and can lead to a more efficient healthcare system.

AI may be able to ease the workload of pathologists, especially in today's healthcare climate where there is a worldwide shortage of healthcare workers, he noted. "This is a first step, but I think it will speed things up a lot."

"We can assess thousands of proteins in the blood, but what does it mean? How does that translate into a decision by a doctor? This is when we need artificial intelligence," Krag added.

Comment