How palevas/syntax-score-prediction works
This repository implements an end-to-end video-based SYNTAX score prediction benchmark. Its workflow trains an artery-specific video backbone, uses that backbone to initialize a study-level sequence model, then applies sequence models to left and right coronary artery videos and combines their estimates into a study-level prediction. The dataset classes read study metadata and DICOM video. README and sampled source establish the training and inference paths; other benchmark capabilities, such as model variants, are retained, but unsampled internals are not expanded.
Main entry points: seq_apply.py Inference runner · seq_train.py Sequence training · backbone_train.py Backbone training · seq_dataset.py Study video dataset
Components 13
Clip backbone
- Clip dataset
backbone_dataset.py - Backbone training
backbone_train.py - Video classifier
backbone_model.py - Backbone weights
Study sequence
- Study video dataset
seq_dataset.py - Sequence training
seq_train.py - Sequence predictor
seq_model.py - Sequence weights
Prediction
- Inference runner
seq_apply.py - Study predictions
- Evaluation metrics
seq_apply.py
Other components
- Researcher
- CardioSYNTAX data
Connections 18
- Researcher → Backbone training: trains backbone
README.md - CardioSYNTAX data → Clip dataset: provides metadata and video
backbone_dataset.py - Backbone training → Clip dataset: loads batches
backbone_train.py - Backbone training → Video classifier: trains model
backbone_train.py - Backbone training → Backbone weights: saves weights
backbone_train.py - Researcher → Sequence training: trains sequence model
README.md - CardioSYNTAX data → Study video dataset: provides study videos
seq_dataset.py - Sequence training → Study video dataset: loads batches
seq_train.py - Sequence training → Sequence predictor: trains model
seq_train.py - Sequence training → Backbone weights: loads initialization
seq_train.py - Sequence training → Sequence weights: saves checkpoints
seq_train.py - Researcher → Inference runner: runs inference
README.md - CardioSYNTAX data → Inference runner: provides study metadata
seq_apply.py - Inference runner → Study video dataset: loads videos
seq_apply.py - Inference runner → Sequence predictor: runs prediction
seq_apply.py - Sequence weights → Inference runner: provides model weights
seq_apply.py - Inference runner → Study predictions: writes study scores
seq_apply.py - Inference runner → Evaluation metrics: computes metrics
seq_apply.py