SLIP: Interactive 3D Segmentation Needs Latency and Undo
SLIP caches 3D patch features and undoes clicks with 140 MB deltas. Cross-dataset click latency is 0.06 s versus 0.31 s for nnInteractive. MRI uterus still goes to nnInteractive.
CT-DeltaBench: Prior vs Current CT Comparison Is Still Unsolved
CT-ΔBench scores prior-vs-current CT difference reports on 2,638/169 CT-RATE pairs. Zero-shot Change-F1 tops out at 0.0175; DeltaMed full SFT reaches 0.1980.
AMPLIFAI: Feature-Level Masks on Multiphase CT for LI-RADS Liver AI
AMPLIFAI adds LI-RADS categories plus voxel masks for APHE, washout, and capsule on 590 multiphase CTs, so liver AI can show the features in a viewer.
RadYOLO: Why Fast 3D Detection, Not Just Dense Segmentation, Changes What Radiology AI Can Ship
RadYOLO is a 3D YOLO11 for joint detection and instance segmentation on CT and MRI. It is 8-46x faster than nnU-Net on GPU and stronger on lesions.
PatchChestCT: Why Spatial Labels, Not Image-Level Findings, Are the Missing Piece for Chest CT AI
PatchChestCT adds physician-reviewed 3D patch labels for nine chest CT findings on 2,201 CT-RATE studies, so models can localize, not just classify.
How to Convert a NIfTI File into a DICOM Series Using Python
This blog is about how to use the nifti2dicom functin to convert nifti file or files into dicom series.
RADAR: a 3D radiology report-review benchmark
RADAR is a new multimodal benchmark testing whether AI can review and improve draft radiology reports against 3D imaging studies — and why it matters for medical imaging product teams.
When Reading the Scan and Writing the Report Become One Step | AI Radiology Reporting
A session-based radiology reporting workflow: read naturally in the viewer, narrate findings, and turn the review into a structured draft report.
Merlin: CT-native foundation models for radiology AI
Merlin shows how CT-native vision-language foundation models could shape radiology AI, with released code, model weights, and dataset access for researchers.
GPU volume rendering for faster CT visualization
Using GPU volume rendering with vtkGPUVolumeRayCastMapper made our CT viewer dramatically faster—smooth navigation, instant cropping, and better visuals even on large scans.