How benkeel/lymph_node_classification_miua works
Project type and purpose
• This is a research-style machine-learning pipeline for “interpretable prediction of lymph node metastasis in rectal cancer MRI” using a variational autoencoder (VAE) plus a multilayer perceptron (MLP) classifier.
• The codebase includes data preprocessing notebooks, model definitions, training/evaluation utilities, hyperparameter-sweep scripts, and a results directory.
File structure
• Top-level folders:
• Data_Preprocessing – Jupyter notebooks for exploratory data inspection, image extraction from raw MRI, and dataset preprocessing.
• VAE-MLP – the core modeling package, split into:
– models: VAE_2D_model.py and MLP_model.py
– utils: dataset loaders, loss functions, training/test routines, helper code
– scripts: a set of .ipynb notebooks for cross-validation, hyperparameter sweeps, Grad-CAM, latent-space analyses, ablation studies, etc.
• Results – holds configuration files, training logs, model outputs, plots, and error analyses for each experimental setup (CNN, VAE, VAE-MLP, cross-validation).
• requirements.txt – defines dependencies (PyTorch, scikit-learn, matplotlib, etc.).
README for additional insights
• The README confirms the pipeline stages (preprocessing → model training → evaluation) and lists the experiments and outputs under Results.