The Lung Cancer Channel is supported with funding from Johnson & Johnson (Gold) and Takeda (Gold).
VJOncology 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.
WCLC 2025 | Challenges in translating AI-based predictive models into clinical practice
Nicolas Alcala, International Agency for Research on Cancer, Lyon, France, highlights the need for a robust proof of concept and large-scale data collection to translate artificial intelligence (AI)-based predictive models into routine oncology practice. Dr Alcala emphasizes the importance of multi-modal data, including sequencing and imaging data, to understand tumor biology and develop reproducible models. This interview took place at 2025 World Conference on Lung Cancer (WCLC) in Barcelona, 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
So I think there are several. There are many, many steps that need to be taken. So first, I think we need to have a very good proof of concept. So in the research setting, we need to prove that it works and we can estimate something which is useful to discriminate at least the biology of the tumors and understand this using AI. And for this, we need probably first a lot more data...
So I think there are several. There are many, many steps that need to be taken. So first, I think we need to have a very good proof of concept. So in the research setting, we need to prove that it works and we can estimate something which is useful to discriminate at least the biology of the tumors and understand this using AI. And for this, we need probably first a lot more data. We need multi-modal data where we have for the same tumors, we have like the sequencing data and also the images scanned. And we need this for a decently large amount of tumors. We need it from different samples to test also how reproducible this can be, just for training the models and testing the models. Otherwise, we are a bit blind to what happens. So it’s also for me a call to the community to please be open and we need to share this information if we need to advance and we need to benefit the patients. In clinical settings, how it works, how it manages to stratify the different patients into different groups. And we need this to be tested also in clinical trials to see if when we integrate this, we can identify the subset of patients that respond to this or that other drug.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
Read more...
