PRANA is an AI-assisted clinical decision support system designed to streamline the liver transplant allocation process. It integrates donor and recipient management, longitudinal clinical monitoring, AI-based survival prediction, dynamic waitlist ranking, compatibility matching, and organ allocation into a unified platform.
An end-to-end AI-assisted liver transplant allocation platform integrating longitudinal clinical monitoring, deep learning-based survival prediction, dynamic waitlist prioritization, and compatibility-driven organ allocation.
Note: PRANA is developed as an academic/hackathon project and is intended for educational and research purposes only. It is not a replacement for clinical judgment or certified medical software.
- Problem Statement
- Key Contributions
- Features
- System Architecture
- Database Design
- Tech Stack
- AI Workflow
- Repository Structure
- Screenshots
- Installation
- Model Availability
- Future Improvements
- Disclaimer
- Authors
- License
Liver transplantation requires careful coordination between donor registration, recipient evaluation, clinical scoring, compatibility assessment, and waitlist prioritization.
Traditional workflows often involve multiple disconnected systems, making the process time-consuming and prone to delays.
PRANA addresses this by providing a centralized platform with AI-assisted decision support while ensuring that final allocation decisions remain under physician supervision.
- End-to-end liver transplant allocation platform built with Django
- AI-assisted 90-day survival prediction using a BiLSTM + Attention architecture
- Longitudinal patient monitoring using 15-day laboratory trends
- Dynamic waitlist generation and priority ranking
- Compatibility-based donor-recipient matching
- Automated allocation workflow with notifications
- Interactive dashboard for transplant management
- Register liver donors
- Brain death verification
- Next-of-kin consent verification
- Automatic donor ID generation
- Donor status tracking
- Recipient registration
- Longitudinal laboratory monitoring
- Clinical assessment records
- Automatic recipient ID generation
- Patient progress tracking
Automatically computes multiple liver disease severity scores including:
- MELD
- MELD-Na
- MELD 3.0
- Child-Pugh (CTP)
- ALBI
- APRI
- FIB-4
- 90-day post-transplant survival prediction
- Mortality probability estimation
- Risk categorization
- AI-based ranking recommendation
- Doctor verification workflow
- Automatic waitlist generation
- Priority score calculation
- Dynamic patient ranking
- Active waitlist management
The matching engine evaluates:
- Blood group compatibility
- Height compatibility
- Weight compatibility
- Standard Liver Volume (SLV)
- Graft-to-Recipient Weight Ratio (GRWR)
Only medically compatible recipients proceed to allocation.
Automatically:
- Selects the highest-ranked compatible recipient
- Creates allocation records
- Updates donor status
- Updates recipient status
- Removes recipient from active waitlist
- Generates hospital notifications
- Allocation notifications
- Read/unread tracking
- Notification history
Provides real-time overview of:
- Registered donors
- Registered recipients
- Active waitlist
- AI predictions
- Organ allocations
- High-priority patients
- Notifications
Hospital
│
▼
Dashboard
│
├──────────────┐
│ │
▼ ▼
Donor Module Recipient Module
│ │
│ ▼
│ Clinical Assessment
│ │
│ AI Prediction
│ │
└────────► Waitlist
│
▼
Matching Engine
│
▼
Allocation Engine
│
▼
Notification System
Main entities:
- Hospital
- Liver Donor
- Recipient
- Recipient Lab
- Clinical Assessment
- Prediction
- Waitlist
- Allocation
- Notification
Relationships are managed using Django ORM with ForeignKey and OneToOne relationships.
- Python
- Django
- SQLite
- Django ORM
The survival prediction model is built using a deep learning architecture designed for longitudinal clinical data.
- Bidirectional LSTM (BiLSTM)
- Attention Mechanism
- Fully Connected Layers
- Sigmoid Output Layer
The model analyzes temporal trends in patient laboratory values collected over 15 days and predicts:
- 90-day post-transplant survival probability
- Mortality probability
- Risk category
The prediction is then incorporated into the dynamic waitlist ranking, subject to physician verification.
The AI prediction module utilizes a Bidirectional Long Short-Term Memory (BiLSTM) network enhanced with an Attention Mechanism.
Clinical Time-Series
(15 Days)
│
▼
BiLSTM Encoder
│
▼
Attention Layer
│
▼
Dense Layers
│
▼
Sigmoid
│
▼
Survival Probability
The BiLSTM captures temporal dependencies from both past and future laboratory observations, while the Attention layer emphasizes clinically significant time points before generating the final prediction.
- HTML
- CSS
- Bootstrap
- JavaScript
Recipient Registration
│
▼
Laboratory Data Collection
│
▼
Clinical Score Calculation
│
▼
AI Survival Prediction
│
▼
Doctor Verification
│
▼
Waitlist Ranking
│
▼
Compatibility Matching
│
▼
Organ Allocation
| .gitignore
| manage.py
| readme.md
| requirements.txt
|
+---liver
| | admin.py
| | apps.py
| | calculations.py
| | context_processors.py
| | demo_data.py
| | forms.py
| | models.py
| | signals.py
| | tests.py
| | urls.py
| | utils.py
| | views.py
| | __init__.py
| |
| +---ai
| | | model.py
| | | model_loader.py
| | | predictor.py
| | | preprocessing.py
| | | risk.py
| | |
| | +---artifacts (Model Assets excluded from Public Repo)
| |
| +---services
| | | allocation.py
| | | matching.py
| | | prediction.py
| | | waitlist.py
| | |
| | \---__pycache__
| | allocation.cpython-313.pyc
| | matching.cpython-313.pyc
| | prediction.cpython-313.pyc
| | waitlist.cpython-313.pyc
| |
| +---static
| | +---css
| | | style.css
| | |
| | \---js
| +---templates
| | \---liver
| | add_lab_report.html
| | allocation.html
| | allocation_detail.html
| | allocation_list.html
| | base.html
| | dashboard.html
| | donor_list.html
| | donor_register.html
| | history.html
| | home.html
| | landing.html
| | login.html
| | notification_detail.html
| | notification_list.html
| | recipient_detail.html
| | recipient_list.html
| | recipient_predictions.html
| | recipient_register.html
| | support.html
| | waitlist.html
|
|
+---PRANA
| | asgi.py
| | settings.py
| | urls.py
| | wsgi.py
| | __init__.py
| |
|
\---screenshots
AI_Survival_Prediction.png
Dashboard_1.png
Dashboard_2.png
Liver_Allocation.png
Liver_Allocation_2.png
Patient_details_1.png
Patient_Details_2.png
Patient_Details_3.png
Recipient_Register(1).png
Recipient_Register.png
Waitlist.png
.png)
Clone the repository
git clone https://github.com/yourusername/PRANA.gitNavigate to the project
cd PRANAInstall dependencies
pip install -r requirements.txtpython manage.py makemigrationspython manage.py migrateStart the server
python manage.py runserverOpen
http://127.0.0.1:8000/
Note: AI model artifacts and preprocessing files are intentionally omitted from this public repository. Certain AI prediction features will require these assets during runtime.
The repository contains the complete source code for the PRANA platform.
To protect ongoing research and responsibly distribute AI assets, the following components are intentionally excluded from the public repository:
- Trained model weights (.pth)
- Preprocessing pipelines (.pkl)
- StandardScaler parameters
- Training datasets
- Intermediate checkpoints
The application architecture, workflow, and source code are fully available. Model artifacts can be demonstrated during evaluation or provided separately if required by the organizers.
- Multi-organ transplant support
- Hospital authentication
- Integration with Electronic Medical Records (EMR)
- Explainable AI (XAI)
- National transplant registry integration
- REST API for external hospital systems
This project was developed as an academic/hackathon prototype to demonstrate the integration of artificial intelligence into transplant allocation workflows.
The AI model is intended to assist clinicians and does not replace medical decision-making.
Yash Singh
Computer Science & Engineering (AI & ML)
This repository is intended for educational, research, and hackathon evaluation purposes.
Please contact the authors before redistributing trained models or research assets.