ASCO 2025 | Exploring the use of a novel deep learning-based MRI to better identify ICANS after CAR-T

So this year at ASCO, we presented our work on immune effector cell-associated neurotoxicity. So ICANS, it’s a really devastating syndrome. You can see it in up to 70% of patients who get CAR T-cell therapy. It can be devastating to the point of coma, death. I mean, there’s a variety of ways patients can present neurologically. As of now, we have several established risk factors in the literature, like high LDH, high disease burden. But from a neuroimaging perspective, we’re lacking a bit in terms of any sort of predictive neuroimaging predictive biomarkers of developing this syndrome, or a clear neuroimaging diagnostic criteria for establishing this diagnosis when patients actually have CAR-T-associated neurotoxicity. So what we did is we looked at all of the patients at UCSD. There’s 163 of them who had non-Hodgkin’s lymphoma or ALL treated with CAR-T with commercial products. About half of them developed ICANS. About half of those were grade one to two. Half of them are grade three to four. What we did is we looked at all of those different labs that they had available. Again, this is retrospective. So whatever labs they had available prior to CAR-T. And then we looked at their clinical demographic characteristics, prior treatments they’d had. We looked at all patients at three days after CAR-T infusion. And then those who ended up getting ICANs, we looked at those same features during their presentation of ICANS. And if they did not get ICANs, we looked at them at seven days, which is our median time to development of ICANs. So we looked at these different time points to look at imaging features as well as serum, so blood-based biomarkers, labs that we had available, and then also clinical presentation. So the take-home was really that prior to CAR T-cell therapy, we saw that intrathecal chemotherapy actually increased the risk of developing ICANS and then also elevated LDH, which was previously established in the literature. We saw that cytokine release syndrome severity was associated with ICANS severity. And then what I was most excited about, and hence the word novel in the title, which makes it more exciting always, is that we used deep learning-based MRI to try to auto-segment features that are associated with ICANS. So I work with a brilliant radiologist, Jeff Rudy, at UCSD, and he’s developed a deep learning-based algorithm that he’s applied for other neurologic conditions like glioma and brain metastases. And basically, we’re developing this to be applied for ICANS in order to identify again and segment out features that can be associated with development of ICANS. So we looked at 93 MRIs after CAR T-cell therapy. It’s over 21 patients, so not a huge portion, but there were some patients we could look at out of those 163. And even in that small subgroup, we did see that patients who had clinical diagnosis of ICANS actually had significantly greater deep learning-based MRI-segmented T2 FLAIR hyperintensity, so MRI changes of white matter inflammation, compared to patients who did not have ICANS. And so while it’s all very exciting preliminary data, it really just speaks to the fact that I think we need more regimented neuroimaging to really answer this question and come up with a clear, comprehensive phenotype of both serum and neuroimaging features of ICANs. And I’m very excited that now we’re starting to actually, we’re in the process through the startup process of actually opening a trial at UC San Diego. I received some funding from the National Cancer Center Network Young Investigator Award to support this work to answer this question in order to get MRIs before CAR-T and after CAR-T at different time points to apply this deep learning-based algorithm, better train it, and for us to really answer this question of who is at risk for getting ICANS and how can we identify it early and diagnose it more accurately.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.