Note The end-to-end workflow as a notebook for Google Colab. - Live on Google Colab: https://colab.research.google.com/drive/1qofahQ8LztTrB_Us9j1Iyz2aYeS2_2rH?usp=sharing - GitHub Repo: https://github.com/The-Adimension/ATRIA-EchoTrace - Discussion on Build with Google AI Forum: https://discuss.ai.google.dev/t/atria-echotrace-fine-tuning-medgemma-1-5-for-polygon-based-heart-structure-contouring/172907

The-Adimension/EchoTrace-MedGemma-CAMUS

Image-Text-to-Text • Updated Jul 16 • 1

Note Tuned on ~1600 echocardiographic frames. Pre-processed from CAMUS dataset DICOMs as PNGs. Resolution ~600x500 px. Ground truth tracings translated into 1000x1000 x-y coordinate grid. Dataset Citation: Leclerc et al. (2019). Deep Learning for Segmentation using an Open Large-Scale Dataset in 2D Echocardiography. IEEE Transactions on Medical Imaging. https://doi.org/10.1109/tmi.2019.2900516 — https://www.creatis.insa-lyon.fr/Challenge/camus/databases

The-Adimension/EchoTrace-MedGemma-EchoNet

Image-Text-to-Text • Updated Jul 16 • 1

Note Tuned on ~15,000 echocardiographic frames. Pre-processed from EchoNet-Dynamic dataset AVIs as PNGs. Resolution 112x112 px (upscaled to 224x224 px). Tracing coordinates translated into 1000x1000 x-y coordinate grid. Dataset Citation: Ouyang et al (2019). EchoNet-Dynamic: A Large New Cardiac Motion Video Data Resource for Medical Machine Learning. Stanford AIMI. https://doi.org/10.7171 — https://echonet.github.io/dynamic/

google/medgemma-1.5-4b-it

Image-Text-to-Text • 4B • Updated Apr 13 • 272k • 828

Note Base Model for the ATRIA-EchoTrace Adapters. A Gemma 3-based model with instruction-tuned 4 billion parameter multimodal architecture. Optimised for medical tasks. Model Citation: Sellergren et al. (2026). MedGemma 1.5 Technical Report. arXiv preprint. https://doi.org/10.48550/arXiv.2604.05081 — https://deepmind.google/models/gemma/medgemma/