IEEE ISBI 2022 paper: CEUSegNet: A Cross-Modality Lesion Segmentation Network for Contrast-Enhanced Ultrasound, a cooperation achievement by researchers from Institute of Automation CAS and Lanzhou University Second Hospital.
Modify datadir to your data path first and then run the file train_helper.py.
- Contrast-Enhanced Ultrasound (CEUS) usually presents two modalities on video frames at the same time, i.e. ultrasound part and contrast-enhanced part.
- We can determine a rough location for lesion on ultrasound part and then finely sketch the region of interest on contrast-enhanced part.
- In this way, a video segmentation task can be converted into a frame segmentation task.
Our work can achieve a comparable performance with clinicians on breast lesion and cervical lymphadenopathy segmentation task. More details can refer to our paper.
| Input size | Time (ms) | MACs(G) | Params(M) |
|---|---|---|---|
| 128 * 128 | 20.82±2.62 | 12.74 | 9.281 |
| 375 * 375 (origin) | 68.51±0.41 | 108.51 | 9.281 |
Our code is based on Pytorch-UNet. Thanks!