The edition · Radiology
Nonmass lesions enter the breast ultrasound lexicon
BI-RADS v2025 formally admits lesions with no discrete margin or shape, classified by distribution, with a malignancy rate high enough to warrant tissue diagnosis. Plus the first consensus definitions for sacroiliac joint lesions on CT, what a language model does to a resident's confidence, and the fibroid that should not be morcellated.
The edition in brief
The BI-RADS v2025 manual incorporates nonmass lesions into the breast ultrasound lexicon, classified by distribution as segmental, linear, focal or regional. These lesions lack a discrete margin or shape, carry an overall high malignancy rate warranting tissue diagnosis, and are affected by patient positioning, transducer orientation and operator technique. A modified Delphi process in the European Society of Musculoskeletal Radiology arthritis subcommittee established the first consensus definitions for seven structural sacroiliac joint lesions on CT: erosions, pseudo-widening, subchondral sclerosis, backfill, joint space narrowing, bone bud and ankylosis. Agreement ranged from 76% to 96%. Fat metaplasia was excluded for poor CT detectability. In 93 thoracic imaging cases, a large language model given readers' free-text descriptions reached 63.9% accuracy against 52.7% given images alone. Reader accuracy rose from 56.3% to 65.6% for thoracic radiologists and from 42.4% to 58.5% for residents. Residents accepted the model's favoured diagnosis more often (73.2% vs 48.9%) and switched to an incorrect diagnosis after a misleading output more often (60.6% vs 32.4%). Across a 20-centre network running 10 artificial intelligence tools from 7 vendors, median result latency was 2.06 minutes, 72% of it data routing rather than computation. Results arrived after the report was finalised in 7.2% of cases overall and 13.2% for chest CT. Turnaround time fell 26% for trauma radiography and 18% for knee MRI. A Radiology review reports contrast-enhanced MRI with diffusion-weighted imaging gives 88% to 94.6% accuracy for uterine leiomyosarcoma, against a risk of about one in 770 at surgery for presumed fibroids.
Nonmass lesions are now part of the breast ultrasound lexicon
Report nonmass breast ultrasound findings using the v2025 distribution descriptors and treat them as warranting tissue diagnosis, not surveillance.
Seven agreed definitions for sacroiliac joint lesions on CT
Describe sacroiliac joint findings on CT using the seven agreed terms, including on scans done for other reasons, and keep recommending MRI where early sacroiliitis is suspected.
A language model helped residents most, and misled them most
A text-based language model workflow improved diagnostic accuracy most in residents, who also switched to a wrong answer after a misleading output nearly twice as often as consultants.
The bottleneck was the network, not the algorithm
Measure artificial intelligence latency archive to archive and track how often results arrive after the report is signed — routing, not the algorithm, determined whether these tools were usable.
Work the review areas on every chest radiograph, in the same order
Check technical adequacy first, then run the same review-area sequence after every primary read — most misses are second findings, not subtle ones.
Before a fibroid is morcellated, the MRI question is whether it is a sarcoma
Answer the sarcoma question explicitly in the report before fibroid surgery, and recommend contrast-enhanced MRI with diffusion-weighted imaging where it is unresolved.
Read the rest in the app
You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this one is waiting in it.
Scan to keep reading on your phone. No account needed to start.
Tomorrow morning, before your first patient
One edition a day for radiology, written by the desk, every claim tied to its paper. Six minutes.
Get the app — free