About the Project

MeroBERT-CNN is a deep learning-based Clinical Decision Support System (CDSS) specifically designed for the precise prediction of meropenem trough concentrations in critically ill ICU patients.

1. Model Architecture

This platform utilizes the innovative MeroBERT-CNN fusion network. The model effectively combines the feature extraction capabilities of large-scale pre-trained models with the local feature capture abilities of Convolutional Neural Networks (CNNs):

BERT-based Embedding: Performs deep semantic mapping of complex clinical covariates.
CNN Feature Extractor: Captures non-linear fluctuations in patient physiological indicators.
Optimized Performance: Demonstrates superior robustness compared to traditional PopPK models.

2. Interpretability (XAI)

To ensure clinical transparency, the system integrates the SHAP (SHapley Additive exPlanations) algorithm. Each prediction is accompanied by a feature contribution plot, helping physicians understand how specific factors impact the final estimate.

3. Clinical Guidance Thresholds

The visualization module utilizes thresholds based on the latest clinical pharmacology consensus:

Min Effective (4 mg/L)

Target Efficacy

Ensures drug concentrations remain above the MIC for common pathogens to prevent treatment failure.

Max Alert (32 mg/L)

Toxicity Risk

Exceeding this threshold significantly increases the risk of neurotoxicity (e.g., seizures) in ICU patients.

4. Data Privacy

This system operates on a transient processing basis. All data is processed in-memory and no persistent storage is performed on the server. Users should ensure data is de-identified before upload.

⚠️ Medical Disclaimer:
Prediction results are for academic research and supplementary reference only. All clinical dosing adjustments must be made by a licensed physician.