The Lymphoma Channel is supported with funding from AstraZeneca (Diamond), BMS (Gold), Johnson & Johnson (Gold), Takeda (Silver) and Galapagos (Bronze).
The Multiple Myeloma Channel is supported with funding from Sanofi (Gold) and Legend Biotech (Bronze).
VJHemOnc is an independent medical education platform. Supporters, including channel supporters, have no influence over the production of content. The levels of sponsorship listed are reflective of the amount of funding given to support the channel.
EBMT 2026 | Deep learning-based profiling of CAR-T immune dynamics from routine peripheral blood smears
Roni Shouval, MD, PhD, Memorial Sloan Kettering Cancer Center, New York, NY, discusses the development of a high-throughput deep learning-based framework for profiling of CAR-T immune dynamics from standard peripheral blood smears. Dr Shouval explains that his team developed a deep learning model to automatically identify six reproducible lymphocyte morphologies, achieving good accuracy and identifying certain morphologies that correlate with better progression-free survival (PFS) following CAR T-cell therapy. This interview took place at the 52nd Annual Meeting of the EBMT in Madrid, Spain.
These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.
Transcript
It’s very well established that immune dynamics of CAR T-cells after infusion drive many of the toxicities we see and also resistance. Current tools for immune monitoring rely on intermittent or resource-intensive tools such as digital droplet PCR or flow cytometry. What we did in this work is use image analysis from peripheral blood in patients receiving CAR T-cells and try to leverage that for immune monitoring following CAR-T...
It’s very well established that immune dynamics of CAR T-cells after infusion drive many of the toxicities we see and also resistance. Current tools for immune monitoring rely on intermittent or resource-intensive tools such as digital droplet PCR or flow cytometry. What we did in this work is use image analysis from peripheral blood in patients receiving CAR T-cells and try to leverage that for immune monitoring following CAR-T. Essentially, a group of pathologists referenced or annotated cells, lymphocyte cells, after CAR T-cell infusion, identifying six reproducible morphologies. After we did that, we could identify these references and train and test a deep learning model to automatically identify them and apply them at scale.
We show that this type of model has good accuracy and achieves an AUC of approximately 0.98 and an accuracy of approximately 80%. After developing this model, we could apply it in scale to profile essentially serial lymphocytes or peripheral blood smears of about 600 patients treated at MSK. Next, after we did that, we could see changes in different morphologies of lymphocytes after CAR T-cells, and we were also able to identify certain types of morphologies that correlate with better progression-free survival. And this surpasses tools such as just the absolute lymphocyte count, which was not discriminatory in this cohort. So overall, we think this opens a new avenue for immune monitoring and CAR T-cell therapy, and we hope to continue and develop it and prospectively validate it.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
Read more...
Disclosures
Speaker honoraria: MSD, Incyte, Sanofi; Consultancy honoraria: AROG Pharmaceuticals, CorMedix.
