The Second Reader
The standard of care in oncology trials is often “double reading”—two radiologists checking the same scan. DeepScan V4 aims to democratize that standard, acting as an AI second reader for every single scan that comes through the hospital.
The Sensitivity Test
We tested V4 on a “gold standard” dataset of 500 anonymized Chest CTs (250 normal, 250 with confirmed pathology).
- The Humans:Â Missed 12% of small nodules (<4mm) due to fatigue or “inattentional blindness.”
- DeepScan V4:Â Missed only 0.8%.
Crucially, V4 flagged three “ground glass opacities” that were barely visible to the naked eye, which pathology later confirmed were early-stage adenocarcinomas.
Workflow Impact
AI is useless if it slows you down. DeepScan excels here. It doesn’t pop up annoying windows. Instead, it pre-processes scans before the radiologist opens them. If it finds something, it highlights the study in the worklist with a red tag. This “triage” feature potentially saves lives by ensuring the sickest patients are read first, rather than in chronological order.
Verdict
DeepScan V4 is no longer an “experimental tool.” It is a safety net. In 2025, practicing radiology without this level of AI assistance is starting to look like malpractice.
