Kangning Cui
← Research / Medical AI

Medical AI

Learning from signals
and moving images.

We develop models that make better use of time in medical data: adapting short-window ECG models to longer recordings, and learning cardiac motion from ultrasound without manual phase labels.

Kangning Cui and collaborators · Biomedical signals and image analysis

01 / NeurIPS’26, Poster

Extend ECG models beyond ten seconds

Kangning Cui · Co-author

Overview of extending frozen 10-second ECG foundation models to longer recordings with a lightweight plug-in moduleEnlarge figure
From snapshots to longer recordings. The framework adds temporal modeling to a frozen 10-second ECG backbone, connecting information across the recording rather than simply pooling independent windows.
Problem
An ECG model trained on 10-second clips cannot directly combine information across a longer, variable-length recording.
Method
Keep the pretrained backbone frozen and add a lightweight module for long-sequence processing and temporal relationships.
Result
Experiments across ECG tasks and backbones improve on sliding-window and pooling baselines without retraining the backbone.
Paper Preprint PDF (preprint)
BibTeX
@inproceedings{tang2026extending,
  author = {Tang, Wei and
            Han, Jinpei and
            Cui, Kangning and
            Carletti, Mattia and
            Gustafsson, Fredrik K. and
            Gowda, Shreyank and
            Palo, Patitapaban and
            Thakur, Anshul and
            Clifton, Lei and
            Morel, Jean-michel and
            Chan, Raymond H and
            Clifton, David and
            Gu, Xiao},
  title = {Extending Pretrained 10-Second {ECG} Foundation Models to Longer Horizons},
  booktitle = {The Fortieth Annual Conference on Neural Information Processing Systems},
  year = {2026},
  url = {https://neurips.cc/virtual/2026/poster/150095},
  note = {Accepted conference paper},
}
Full citation and authors

Extending Pretrained 10-Second ECG Foundation Models to Longer Horizons

Wei Tang, Jinpei Han, Kangning Cui, Mattia Carletti, Fredrik K. Gustafsson, Shreyank N Gowda, Patitapaban Palo, Anshul Thakur, Lei Clifton, Jean-Michel Morel, Raymond H. Chan, David A. Clifton, and Xiao Gu

Full publication record

02 / MICCAI’25, Oral

Find cardiac phases without manual labels

Kangning Cui · Co-author

Cardiac motion profiling: an ultrasound reconstruction model separates static structure from motion and identifies end-diastole and end-systole along a latent trajectoryEnlarge figure
Read the heartbeat from motion. The model learns to reconstruct ultrasound frames while separating static structure from motion. The resulting trajectory identifies end-diastole and end-systole.
Problem
Cardiac phase detection usually needs manually marked frames or segmentation annotations.
Method
Learn motion trajectories through video reconstruction, then identify end-diastole and end-systole in that representation.
Result
On EchoNet-Dynamic, mean absolute error is 3 frames for end-diastole and 2 for end-systole. The study also evaluates fetal echocardiography.
Paper Code PDF (preprint)
BibTeX
@inproceedings{10.1007/978-3-032-05185-1_31,
  author = {Yang, Yingyu and
            Yang, Qianye and
            Cui, Kangning and
            Peng, Can and
            D'Alberti, Elena and
            Hernandez-Cruz, Netzahualcoyotl and
            Patey, Olga and
            Papageorghiou, Aris T. and
            Noble, J. Alison},
  title = {Latent Motion Profiling for Annotation-Free Cardiac Phase Detection in Adult and Fetal Echocardiography Videos},
  booktitle = {Medical Image Computing and Computer Assisted Intervention -- {MICCAI} 2025},
  editor = {Gee, James C. and
            Alexander, Daniel C. and
            Hong, Jaesung and
            Iglesias, Juan Eugenio and
            Sudre, Carole H. and
            Venkataraman, Archana and
            Golland, Polina and
            Kim, Jong Hyo and
            Park, Jinah},
  year = {2026},
  pages = {316--325},
  publisher = {Springer Nature Switzerland},
  address = {Cham},
  isbn = {978-3-032-05185-1},
  doi = {10.1007/978-3-032-05185-1_31},
  url = {https://doi.org/10.1007/978-3-032-05185-1_31},
}
Full citation and authors

Latent Motion Profiling for Annotation-free Cardiac Phase Detection in Adult and Fetal Echocardiography Videos

Yingyu Yang, Qianye Yang, Kangning Cui, Can Peng, Elena D'Alberti, Netzahualcoyotl Hernandez-Cruz, Olga Patey, Aris Papageorghiou, and Alison Noble

Full publication record

Complementary work on image quality, tracking, signal analysis, and segmentation.

Browse all Medical AI publications · Related patents

Research and collaboration

For collaboration on biomedical signals, image analysis, or learning with limited annotations, contact Kangning Cui. Current research assistant opportunities and PhD admissions status are on the Team page.