Quantizing reconstruction losses for improving weather data synthesis.

The stochastic synthesis of extreme, rare climate scenarios is vital for risk and resilience models aware of climate change, directly impacting society in different sectors. However, creating high-quality variations of under-represented samples remains a challenge for several generative models. This paper investigates quantizing reconstruction losses for helping variational autoencoders (VAE) better synthesize extreme weather fields from conventional historical training sets. Building on the classical VAE formulation using reconstruction and latent space regularization losses, we propose various histogram-based penalties to the reconstruction loss that explicitly reinforces the model to synthesize under-represented values better. We evaluate our work using precipitation weather fields, where models usually strive to synthesize well extreme precipitation samples. We demonstrate that bringing histogram awareness to the reconstruction loss improves standard VAE performance substantially, especially for extreme weather events.

A bifurcation identifier for IV-OCT using orthogonal least squares and supervised machine learning.

Intravascular optical coherence tomography (IV-OCT) is an in-vivo imaging modality based on the intravascular introduction of a catheter which provides a view of the inner wall of blood vessels with a spatial resolution of 10-20 µm. Recent studies in IV-OCT have demonstrated the importance of the bifurcation regions. Therefore, the development of an automated tool to classify hundreds of coronary OCT frames as bifurcation or nonbifurcation can be an important step to improve automated methods for atherosclerotic plaques quantification, stent analysis and co-registration between different modalities. This paper describes a fully automated method to identify IV-OCT frames in bifurcation regions. The method is divided into lumen detection; feature extraction; and classification, providing a lumen area quantification, geometrical features of the cross-sectional lumen and labeled slices. This classification method is a combination of supervised machine learning algorithms and feature selection using orthogonal least squares methods. Training and tests were performed in sets with a maximum of 1460 human coronary OCT frames. The lumen segmentation achieved a mean difference of lumen area of 0.11 mm(2) compared with manual segmentation, and the AdaBoost classifier presented the best result reaching a F-measure score of 97.5% using 104 features.