dsh-medomni
Medical imaging specialist tools for deepseek-harness: MAIRA-2 and MedGemma report generation for chest X-ray/CT/MRI/retinal (fundus) images, TotalSegmentator organ segmentation, BiomedCLIP zero-shot ultrasound classification, and BiomedParse text-prompted segmentation across X-ray, CT, MRI, ultrasound, and retinal (fundus) images.
安装
npx -p @deepseek-ai/dsh dsh plugin --profile web add github:bowang-lab/dsh-medomni说明文档
阅读完整 README ↗Usage examples
Chest X-ray report:
"Generate a radiology report for this chest X-ray:
/path/to/chest_xray.png"
calls xray_report_medgemma, xray_report_maira, or xray_grounded_report_maira when finding evidence/bounding boxes are useful.
Find something specific, by name, on any of the five modalities:
"Segment the gallstone in this ultrasound image." "Are there any microaneurysms in this fundus photo?"
calls the matching _segmentation_biomedparse tool only when you ask for segmentation, masks, overlays, or localization. BiomedParse takes any free-text finding or anatomical structure, but its mask is localization, not diagnosis.
Disambiguate a vague ultrasound request first:
"What's in this ultrasound before you segment anything?"
calls ultrasound_classify_biomedclip to narrow down anatomy/pathology, then a segmentation tool with the winning label as the prompt.
Whole-body organ segmentation on a CT or MRI volume:
"Segment the liver and kidneys in this CT scan:
/path/to/scan.nii.gz"
calls ct_segmentation_totalseg for the fixed anatomical-structure list, or ct_segmentation_biomedparse if you'd rather name a pathology than an organ.
Prior-vs-current comparison:
"Compare this current chest X-ray to the prior one and describe interval change."
calls xray_longitudinal_comparison with both images.