Add like
Add dislike
Add to saved papers

Marginal Space Deep Learning: Efficient Architecture for Volumetric Image Parsing.

Robust and fast solutions for anatomical object detection and segmentation support the entire clinical workflow from diagnosis, patient stratification, therapy planning, intervention and follow-up. Current state-of-the-art techniques for parsing volumetric medical image data are typically based on machine learning methods that exploit large annotated image databases. There are two main challenges that need to be addressed, these are the efficiency in processing large volumetric input images and the need for strong, representative image features. When the object of interest is parametrized in a high dimensional space, standard volume scanning techniques do not scale up to the enormous number of potential hypotheses and representative image features are subject to significant efforts of manual engineering. We propose a pipeline for object detection and segmentation in the context of volumetric image parsing, solving a two-step learning problem: anatomical pose estimation and boundary delineation. For this task we introduce Marginal Space Deep Learning (MSDL), a novel framework exploiting both the strengths of efficient object parametrization in hierarchical marginal spaces and the automated feature design of Deep Learning (DL) network architectures. Deep learning systems automatically identify, disentangle and learn explanatory attributes directly from low-level image data, however their application in the volumetric setting is limited by the very high complexity of the parametrization. More specifically 9 parameters are necessary to describe a restricted affine transformation in 3D (3 for each location, orientation, and scale) resulting in a prohibitive number of scanning hypotheses, in the order of billions for typical sampling. The mechanism of marginal space learning provides excellent run-time performance by learning classifiers in clustered, high-probability regions in spaces of gradually increasing dimensionality, for example starting from location only (3D) to location and orientation (6D) and full parameter space (9D). Given the structure localization, we estimate the 3D shape through non-rigid, DL-based boundary delineation in an Active Shape Model (ASM) framework. In our system we learn sparse adaptive data sampling patterns which replace manually engineered features by automatically capturing structure in the given data. This is also a type of model simplification, ensuring significant computational improvements and preventing overfitting. Experimental results are presented on detecting and segmenting the aortic valve in ultrasound using an extensive dataset of 2891 volumes from 869 patients, showing significant improvements of up to 45.2% over the current methods. To our knowledge, this is the first successful demonstration of the DL potential to detection and segmentation in full 3D data with parametrized representations.

Full text links

We have located links that may give you full text access.
Can't access the paper?
Try logging in through your university/institutional subscription. For a smoother one-click institutional access experience, please use our mobile app.

For the best experience, use the Read mobile app

Mobile app image

Get seemless 1-tap access through your institution/university

For the best experience, use the Read mobile app

All material on this website is protected by copyright, Copyright © 1994-2024 by WebMD LLC.
This website also contains material copyrighted by 3rd parties.

By using this service, you agree to our terms of use and privacy policy.

Your Privacy Choices Toggle icon

You can now claim free CME credits for this literature searchClaim now

Get seemless 1-tap access through your institution/university

For the best experience, use the Read mobile app