The edition · Radiology
Liver MRI adds to CT before pancreatic cancer surgery, and radiomics models are less reproducible than they look
A meta-analysis of 40 head-to-head studies separates what CT and MRI each do best in pancreatic cancer, an analysis of 50 datasets shows how preprocessing leakage inflates radiomics performance, and ablation for small thyroid cancers trades lower complications for a higher recurrence signal.
The edition in brief
A meta-analysis of 40 direct comparative studies found no significant difference between CT, MRI and endoscopic ultrasound for detecting pancreatic ductal adenocarcinoma. MRI was more sensitive and specific than CT for liver metastases (sensitivity 0.89 versus 0.60), CT was better for vascular invasion, and CT had higher specificity for resectability (0.88 versus 0.63). An analysis of 50 radiomics datasets found that leakage from oversampling before cross-validation inflated AUC by a pooled 0.09, and that near-optimal pipelines disagreed on 20 to 30% of patients. A meta-analysis comparing image-guided ablation with surgery for T1N0M0 papillary thyroid carcinoma found higher odds of complications after surgery (OR 2.75) but lower odds of recurrence (OR 0.64). A multi-site observational study of AI tools in a 20-centre outpatient network found AI results arrived after the report was finalised in 7.2% of examinations, with data routing accounting for most latency, and report turnaround was shorter for trauma radiography and knee MRI.
Image-guided ablation versus surgery for small papillary thyroid cancer: fewer complications, more recurrence
Where ablation is offered for a small low-risk papillary carcinoma, discuss both the lower morbidity and the higher recurrence signal, and decide in a multidisciplinary meeting.
Radiomics performance is inflated by leakage and varies with pipeline choice
When reading a radiomics study, check that oversampling and feature steps were done inside cross-validation and that an external test set was used.
AI in a 20-centre radiology network: latency, not the algorithm, limited usefulness
Before buying or judging an AI tool, measure how often its result arrives before the report is signed.
State what the clinician is deciding when you request or report a follow-up scan
Write the clinical question on every request and give a specific interval and modality in every follow-up recommendation.
Pancreatic cancer staging: liver MRI for small metastases, CT for vessels and resectability
Consider adding liver MRI to CT for patients heading to potentially curative pancreatic surgery, and use CT for vascular and resectability assessment.
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