Deep Learning with Anatomical Priors: Imitating Enhanced Autoencoders in Latent Space for Improved Pelvic Bone Segmentation in MRI

TL;DR

This paper introduces a novel deep learning architecture that integrates anatomical priors through latent space imitation of autoencoders, enhancing pelvic bone segmentation accuracy in MRI compared to standard methods.

Contribution

It presents a new encoder-decoder model that incorporates hierarchical features and anatomical priors, improving segmentation performance in medical imaging.

Findings

Improved segmentation accuracy over standard U-Net.

Effective integration of anatomical priors in deep learning.

End-to-end trainable architecture.

Abstract

We propose a 2D Encoder-Decoder based deep learning architecture for semantic segmentation, that incorporates anatomical priors by imitating the encoder component of an autoencoder in latent space. The autoencoder is additionally enhanced by means of hierarchical features, extracted by an U-Net module. Our suggested architecture is trained in an end-to-end manner and is evaluated on the example of pelvic bone segmentation in MRI. A comparison to the standard U-Net architecture shows promising improvements.