Posts

Real‑world validation still matters for imaging AI | Martin Willemink

Image
  Over the past few years, intracranial hemorrhage (ICH) detection has become one of the most mature and regulated use cases in clinical imaging AI. Multiple FDA‑cleared models are now clinically used in radiology workflows. However, recent evidence suggests that deployment maturity does not necessarily mean that models are clinically robust. Since ICH detection has been on the market for a while now, multiple studies on post-deployment evaluation are coming out. Two recent papers examining commercial ICH detection models illustrate both progress and persistent gaps. Together, they emphasize an important lesson for imaging AI research: performance claims are dependent on the data distributions on which models are trained and evaluated. Training an AI model on population A does not mean it will perform well in population B. Understanding where and why models fail is the foundation for the next generation of clinically meaningful AI. Real‑world performance of a commercial ICH model C...
Post a Comment
Read more

Multimodal Data Pipelines: The New Gold Standard in Pharma Research | Segmed Expert

Image
  In the current era, the pharmaceutical industry has largely embraced the “Multimodal Dataset”. R&D teams are successfully integrating genomics, transcriptomics, and Electronic Health Records (EHR) to build longitudinal patient views. This shift toward integrated evidence is driven by the realization that biology is too complex for unimodal analysis. This has made multimodal datasets not just desirable, but essential. What makes multimodal pipelines the new gold standard is not volume alone, it is coherence. When data sources are systematically linked, time aligned, and governed under consistent quality and compliance frameworks, they enable answering research questions that were previously impossible to answer. Questions about patient heterogeneity, real world treatment effectiveness, and disease evolution across care pathways are now becoming accessible. However, a significant gap remains: medical imaging is the most data-rich phenotypic record and is still being treated as ...
Post a Comment
Location: United States
Read more

How Real-World Imaging Data Can Close the AI Validation Gap | Segmed Experts

Image
  Artificial intelligence (AI) has demonstrated impressive performance across a range of medical imaging tasks, from lesion detection and segmentation to prognostication and treatment response assessment. However, most imaging AI systems are still trained and validated using limited and highly curated datasets. Such datasets are often drawn from a small number of institutions, scanners, and patient populations. While these datasets are essential throughout the algorithm development process, they rarely reflect the complexity, heterogeneity, and unpredictability of real-world clinical practice.‍ High performance on internal validation does not guarantee reliable performance in real-world clinical settings. This disconnect is often referred to as the  AI validation gap . It has become one of the central barriers to trustworthy clinical AI. Closing this gap requires validating models against real-world imaging data that reflect true deployment conditions, enabling robust, fair, a...
Post a Comment
Location: United States
Read more