🔵 RESEARCH
Deep learning model predicts breast cancer Ki-67 noninvasively
European Radiology published a multicenter retrospective study on a deep learning model that predicts Ki-67 expression and prognostic stratification in breast cancer without biopsy.
If validated prospectively, biopsy-avoidance AI threatens tissue diagnostics revenue and hands radiology AI vendors leverage in payer negotiations. But retrospective design alone won't convince oncologists to treat without tissue confirmation of multiple biomarkers.
🔵 RESEARCH
JUM issues peritumoral elastography interpretation guidance
Journal of Ultrasound in Medicine published technical and pathologic considerations for interpreting peritumoral elastography, aimed at standardising practice across operators.
Standardised protocols become competitive moats — ultrasound OEMs that embed these interpretation rules into AI workflows can differentiate against basic systems. Following last week's ASE artifact guideline, expect vendors to productise consensus papers quickly.
Journal of Ultrasound in Medicine · 2026-05-06
Read more → (paywall)
🔵 RESEARCH
Trial tests intraoral ultrasound probe for periodontal diagnosis
A clinical investigation is evaluating agreement between high-frequency intraoral ultrasound and gold-standard periodontal probing for measuring pocket depth in early periodontal disease.
If accuracy holds up, intraoral ultrasound opens a non-radiation pathway into the dental imaging market dominated by X-ray. Following 3Shape's recent FDA dental clearance, ultrasound majors may eye dental point-of-care as their next adjacency.
🔵 RESEARCH
Meta-analysis benchmarks deep learning against radiologists in DBT
JMIR published a systematic review and meta-analysis comparing deep learning algorithms with radiologists for breast cancer detection on digital breast tomosynthesis.
Pooled validation arms vendors like Hologic and DBT-focused AI players with reimbursement-grade evidence, likely accelerating M&A. But performance studies say nothing about PACS integration or liability protocols — the actual blockers to deployment.
Journal of medical Internet research · 2026-05-06
Read more →
🔵 RESEARCH
AI chest X-ray tool tested for lung cancer screening recruitment
A clinical trial is evaluating whether an AI tool flagging high 6-year lung cancer risk on chest X-rays improves participation in lung cancer screening CT.
Stratification AI captures screening value without new hardware, a clean software-only commercial model if CMS plays along. The operational test: does pre-screening accelerate CT referral, or just add a decision point to an already fragmented workflow?
🔵 RESEARCH
Prospective CT study compares supine versus prone ILA quantification
European Radiology published a prospective study evaluating positional variability in quantitative CT measurements of interstitial lung abnormalities using same-day supine and prone scans.
Positional effects matter for qCT reproducibility, but prone scanning adds time and throughput friction most departments won't accept routinely. Adoption hinges on whether quantified differences actually change ILD management — not just measurement values.
🩺 EDITOR’S TAKE — FOR CLINICIANS
Watch the biopsy-avoidance thesis carefully — Ki-67 prediction and DBT meta-analyses look strong on paper, but neither replaces tissue confirmation or your sign-off today. Treat these as triage aids, not decision substitutes.
📊 EDITOR’S TAKE — FOR INVESTORS
Imaging AI is moving up the value chain from detection to biomarker prediction and screening recruitment. Software-only plays targeting biopsy avoidance and screening uptake have the cleanest reimbursement story; expect tissue diagnostics incumbents to defend via M&A.
🏭 EDITOR’S TAKE — FOR INDUSTRY
Standardisation papers and meta-analyses are becoming commercial assets — vendors that embed JUM elastography protocols or cite DBT validation data will win procurement conversations. Workflow integration, not algorithm performance, remains the deployment bottleneck.