Clinical research summaries, radiology AI findings, and regulatory updates from the Diagnosify team.
A prospective multi-center trial across eight academic medical centers demonstrates Diagnosify's chest X-ray model achieves 97% sensitivity for pneumonia detection with 91.4% specificity on a diverse patient cohort of 18,400 studies.
Read the SummaryA 6.2mm subpleural nodule flagged by our model after three board-certified radiologists signed off on consecutive reads — a case study in cognitive load, perceptual error patterns, and where AI second readers add measurable value.
Read the ArticleA systematic review of 34 peer-reviewed studies comparing AI-assisted versus unassisted radiologist reads, examining where AI adds measurable value, where it does not, and what questions remain open in the published literature.
Read the SummaryThe official guidance says 90 days. Our first clearance took 14 months. A candid breakdown of predicate selection errors, Additional Information requests, PCCP requirements, and what post-market surveillance actually obligates you to do.
Read the ArticleA detailed account of Diagnosify's 510(k) submission process — from predicate device selection and substantial equivalence arguments to clinical performance testing protocols and the final decision letter from the FDA Division of Radiological Health.
Read the SummaryAI-assisted worklist triage cut median critical-study TAT from 68 minutes to 19 minutes at a Level I trauma center. Data from three deployment sites — a trauma center, community hospital, and teleradiology practice.
Read the ArticleThe United States faces a projected shortage of 42,000 radiologist FTEs by 2035. A clear-eyed analysis of where AI creates genuine clinical capacity — and the cases where it doesn't.
Read the ArticleA 5.7-point sensitivity gap between white and Black patients in our chest X-ray pneumonia model — what caused it, the interventions that narrowed it to 1.2 points, and why demographic performance analysis should be mandatory.
Read the ArticlePACS compatibility, PHI handling, uptime SLAs, HITRUST reviews — the real objections from 14 hospital IT security assessments and what actually resolves them.
Read the ArticleSensitivity and specificity figures across six detection tasks — with the validation set design, reference standard, and case-mix context that make cross-modality comparisons meaningful rather than misleading.
Read the ArticleCommunity hospitals face different constraints than academic centers. A realistic ROI analysis covering software costs, teleradiology savings, locums reduction scenarios, and the volume thresholds where deployment makes financial sense.
Read the ArticleWhat happens between a high-performing peer-reviewed model and a working clinical deployment — distribution shift, PACS integration failures, clinical governance, FDA review, and threshold calibration in sequence.
Read the ArticleAlert fatigue, case-mix drift, normal variant misclassification, and the counterintuitive finding that fewer alerts sometimes means better clinical outcomes — from a 50,000-study post-market surveillance dataset.
Read the Article