EchoGemma

Multimodal echocardiography report generation from DICOM studies. EchoGemma combines an EchoPrime video encoder and a LoRA-fine-tuned MedGemma language model to process full echocardiographic studies and generate clinical text reports.

Input

A folder of DICOM echocardiography video files (a complete study). The model processes all video clips, extracts embeddings and view classifications, then generates a structured clinical report.

Output

A structured echocardiography text report.

Requirements

• Python >= 3.10
• PyTorch 2.10+
• CUDA-capable GPU (recommended)
• ~18 GB disk space for model weights