Blog

Large-scale pre-trained models are reshaping radiology, pathology, and genomics. We explore how biomedical foundation models differ from general-purpose AI and what it takes to deploy them clinically.

Immune checkpoint inhibitor-induced pneumonitis can be life-threatening. Deep learning models trained on baseline CT scans can now identify at-risk patients before toxicity strikes.

LLMs are moving from chatbots to clinical tools. This post examines how models fine-tuned on oncology reports can extract biomarker statuses, treatment responses, and staging information automatically.

CLIP-style and GPT-4V-type models can now align CT scans with clinical text. We break down the architecture, training strategies, and open challenges for VLMs in medical imaging.

PD-L1 testing requires invasive tissue sampling and misses tumor heterogeneity. Deep learning radiomics pipelines trained on multi-center CT cohorts offer a non-invasive alternative for patient selection.

Prognostic biomarker identification is being revolutionized by multimodal AI that jointly processes CT radiomics, RNA-seq, and EHR data. We survey the landscape and key methodological challenges.