beta-VAE for liver

A implementation of beta-VAE and VAE(beta = 1) for Level Set Distribtuion Model(LSDM) of liver

"Construction and Evaluation of a Statistical Shape Model of Liver Using Variational Autoencode", Zhihui Lu et al. 2019 IEEE International Symposium on Biomedical Imaging (ISBI 2019)

Requirements

1. TensorFlow >= 1.4.0
2. SimpleITK
3. tqdm
4. matplotlib
5. scipy

Results

The performance of Model was evaluated with generalization and specificity indices.

We compared our model to the conventional LSDM based on PCA
Generalization and Specificity

Latent Space (blue: training data, orange/pink: test data)

Usage

Input: TFRecord file
Training Model: trainer.py
Evaluate Model: evaluations.py
Plot Latent Space: plot_latent_space.py
Reconstructing Image: predict_gen.py
Generating Image: predict_spe.py

Reference

[1] Kingma, D. P., & Welling, M. (2013). Auto-Encoding Variational Bayes, (Ml), 1–14. https://doi.org/10.1051/0004-6361/201527329

[2] Thiagarajan, B. G., Member, A., & Voyiadjis, G. Z. (2016). Β-Vae: Learning Basic Visual Concepts With a Constrained Variational Framework. Iclr 2017, (July), 1–13.