MACE-MAIS

Official PyTorch implementation of
"An Interpretable Multimodal AI System for Predicting Major Adverse Cardiovascular Events from Comprehensive Patient Profiles "

---

🚀 Overview

MACE-MAIS is an end-to-end interpretable multimodal AI system for predicting Major Adverse Cardiovascular Events (MACE). It integrates:

• Cardiovascular Magnetic Resonance (CMR) imaging
• Electronic Health Records (EHR)

Key features:

• Handles missing modalities robustly
• Provides clinically meaningful explanations

---

🔧 Setup

📋 Prerequisites

• Python ≥ 3.9
• NVIDIA GPU + CUDA (optional, but recommended)

🛠 Installation

# 1. Clone the repository
git clone https://github.com/shaohao011/MACE-MAIS.git
cd MACE-MAIS

# 2. Create and activate conda environment
conda create -n mace-mais python=3.9
conda activate mace-mais

# 3. Install dependencies
pip install -r requirements.txt

---

📂 Usage Guide

1️⃣ Data Preparation

Place your input data under the data/ directory. Follow the format specified in the preprocessing scripts.

# Split survival intervals
python survival_dst_make.py

---

2️⃣ Reasoning Model (LRM)

# Generate SFT data for LRM
python utils/gen_mace_cot.py

# Train LLM using LLaMA-Factory
cd LLaMA-Factory
bash train_llama.sh

---

3️⃣ CMR Image Pretraining

# Step 1: Run pretraining
cd Pre-train
bash do_pretrain.sh

# Step 2: Extract embeddings
python Pre-train/utils/get_embedding.py

---

4️⃣ Survival Analysis

# Train and evaluate
bash do_train_survival.sh

# Tip: Set max_epochs=-1 for testing only

---