MedAssist-Pro represents a breakthrough in medical AI technology. In this release, MedAssist-Pro has significantly enhanced its clinical reasoning and diagnostic accuracy by incorporating extensive medical literature and clinical trial data. The model demonstrates state-of-the-art performance across various healthcare benchmarks, including disease diagnosis, drug interaction analysis, and clinical documentation.
Compared to the previous version, MedAssist-Pro shows remarkable improvements in complex medical scenarios. For instance, in the MedQA benchmark, the model's accuracy has increased from 62% in the previous version to 78.5% in the current version. This advancement stems from enhanced medical knowledge integration: the model now processes an average of 18K tokens per clinical case, compared to 8K tokens in the previous version.
Beyond its improved diagnostic capabilities, this version also offers reduced hallucination rates in medical contexts and enhanced support for multi-modal clinical inputs.
2. Evaluation Results
Comprehensive Medical Benchmark Results
Benchmark
GPT-Med
Claude-Health
MedPaLM-2
MedAssist-Pro
Diagnostic Tasks
Diagnosis Accuracy
0.682
0.695
0.710
0.730
Drug Interaction
0.715
0.728
0.735
0.733
Clinical Reasoning
0.654
0.671
0.689
0.785
Knowledge Tasks
Medical QA
0.621
0.638
0.655
0.647
Radiology Interpretation
0.598
0.612
0.628
0.659
Lab Result Interpretation
0.709
0.722
0.738
0.792
Symptom Analysis
0.687
0.701
0.715
0.731
Clinical Operations
Patient Summarization
0.745
0.761
0.778
0.815
Treatment Recommendation
0.632
0.648
0.665
0.697
Medical Coding
0.698
0.714
0.729
0.718
Surgical Planning
0.578
0.591
0.608
0.597
Safety & Compliance
Patient Triage
0.823
0.838
0.852
0.832
EHR Extraction
0.691
0.705
0.721
0.736
Medical Safety
0.856
0.869
0.882
0.862
Clinical Documentation
0.734
0.749
0.765
0.789
Overall Performance Summary
MedAssist-Pro demonstrates strong performance across all evaluated medical benchmark categories, with particularly notable results in diagnostic tasks and safety compliance.
3. Clinical Interface & API Platform
We offer a clinical interface and API for healthcare professionals to interact with MedAssist-Pro. Please check our official website for more details and HIPAA compliance documentation.
4. How to Run Locally
Please refer to our code repository for more information about running MedAssist-Pro locally.
Compared to previous versions, the usage recommendations for MedAssist-Pro have the following changes:
Medical context system prompt is supported.
It is not required to add special tokens at the beginning of the output to force the model into a specific clinical reasoning pattern.
The model architecture of MedAssist-Pro-Lite is identical to its base model, but it shares the same tokenizer configuration as the main MedAssist-Pro.
System Prompt
We recommend using the following system prompt with clinical context.
You are MedAssist-Pro, a medical AI assistant designed to support healthcare professionals.
Today is {current date}.
IMPORTANT: This AI is for clinical decision support only. Always consult with qualified medical professionals.
Temperature
We recommend setting the temperature parameter $T_{model}$ to 0.3 for clinical applications to ensure consistent and reliable outputs.
Prompts for Clinical Data Processing
For patient record processing, please follow the template to create prompts, where {patient_id}, {record_content} and {clinical_query} are arguments.
clinical_template = \
"""[Patient ID]: {patient_id}
[Clinical Record Begin]
{record_content}
[Clinical Record End]
{clinical_query}"""
For literature-enhanced generation, we recommend the following prompt template where {literature_results}, {cur_date}, and {clinical_question} are arguments.
literature_answer_template = \
'''# The following contents are relevant medical literature:
{literature_results}
In the literature I provide to you, each source is formatted as [source X begin]...[source X end], where X represents the numerical index of each reference. Please cite appropriately using [citation:X] format.
When responding, please keep the following points in mind:
- Today is {cur_date}.
- Evaluate the relevance and quality of each literature source.
- For diagnostic questions, prioritize evidence-based guidelines.
- Always note limitations and recommend appropriate follow-up.
# The clinical question is:
{clinical_question}'''
5. License
This code repository is licensed under the Apache 2.0 License. The use of MedAssist-Pro models is subject to additional healthcare compliance requirements.
6. Contact
If you have any questions, please raise an issue on our GitHub repository or contact us at support@medassist-pro.ai.
