MedGemma-4B ECG Report Generator

This model, developed by Oussama EL, is a 4.3 billion parameter, fully merged and standalone fine-tune of unsloth/medgemma-4b-pt. It specializes in transforming structured ECG findings and patient context into comprehensive, human-readable clinical reports. The fine-tuning utilized the Unsloth library for efficiency, training on 500 curated ECG interpretation examples.

Key Capabilities

• Clinical Report Generation: Synthesizes structured ECG data into coherent reports with impressions, detailed analysis, and clinical recommendations.
• High Clinical Coherence: Achieves a Diagnostic Coherence score of 0.720, indicating strong medical reasoning and logical consistency.
• Robust Operation: Demonstrates a 100% success rate in report generation, ensuring stable and reliable output.
• Linguistic Quality: Scores 30.784 on BLEU-4, reflecting fluent and appropriate medical text generation.
• Efficient Architecture: Optimized for practical deployment with 16-bit merged weights and approximately 8GB GPU memory for inference.

Should I use this for my use case?

This model is ideal for applications requiring automated generation of clinical ECG reports from pre-classified findings. It excels in scenarios where high diagnostic coherence and linguistic quality in medical text are critical. However, it is intended for research and development only and should not replace professional medical advice; all outputs require validation by qualified healthcare professionals.