An official website of the United States government
Abstract Current brain mapping methods highly depend on the regularity, or commonality, of anatomical structure, by forcing the same atlas to be matched to different brains. As a result, individualized structural information can be overlooked. Recently, we conceptualized a new type of cortical folding pattern called the 3-hinge gyrus (3HG), which is defined as the conjunction of gyri coming from three directions. Many studies have confirmed that 3HGs are not only widely existing on different brains, but also possess both common and individual patterns. In this work, we put further effort, based on the identified 3HGs, to establish the correspondences of individual 3HGs. We developed a learning-based embedding framework to encode individual cortical folding patterns into a group of anatomically meaningful embedding vectors (cortex2vector). Each 3HG can be represented as a combination of these embedding vectors via a set of individual specific combining coefficients. In this way, the regularity of folding pattern is encoded into the embedding vectors, while the individual variations are preserved by the multi-hop combination coefficients. Results show that the learned embeddings can simultaneously encode the commonality and individuality of cortical folding patterns, as well as robustly infer the complicated many-to-many anatomical correspondences among different brains.
more »
« less
Automated connectivity-based cortical mapping using registration-constrained classification
An important goal in neuroscience has been to map the surface of the human brain, and many researchers have developed sophisticated methods to parcellate the cortex. However, many of these methods stop short of developing a framework to apply existing cortical maps to new subjects in a consistent fashion. The computationally complex step is often the initial mapping of a large set of brains, and it is inefficient to repeat these processes for every new data sample. In this analysis, we propose the use of a library of training brains to build a statistical model of the parcellated cortical surface and to act as templates for mapping new data. We train classifiers on training data sampled from local neighborhoods on the cortical surface, using features derived from training brain connectivity information, and apply these classifiers to map the surfaces of previously unseen brains. We demonstrate the performance of 3 different classifiers, each trained on 3 different types of training features, to accurately predict the map of new brain surfaces.
more »
« less
- Award ID(s):
- 1734430
- PAR ID:
- 10304184
- Editor(s):
- Gimi, Barjor; Krol, Andrzej
- Date Published:
- Journal Name:
- Medical Imaging 2018: Biomedical Applications in Molecular, Structural, and Functional Imaging
- Volume:
- 10578
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Deep learning has been applied to magnetic resonance imaging (MRI) for a variety of purposes, ranging from the acceleration of image acquisition and image denoising to tissue segmentation and disease diagnosis. Convolutional neural networks have been particularly useful for analyzing MRI data due to the regularly sampled spatial and temporal nature of the data. However, advances in the field of brain imaging have led to network- and surface-based analyses that are often better represented in the graph domain. In this analysis, we propose a general purpose cortical segmentation method that, given resting-state connectivity features readily computed during conventional MRI pre-processing and a set of corresponding training labels, can generate cortical parcellations for new MRI data. We applied recent advances in the field of graph neural networks to the problem of cortical surface segmentation, using resting-state connectivity to learn discrete maps of the human neocortex. We found that graph neural networks accurately learn low-dimensional representations of functional brain connectivity that can be naturally extended to map the cortices of new datasets. After optimizing over algorithm type, network architecture, and training features, our approach yielded mean classification accuracies of 79.91% relative to a previously published parcellation. We describe how some hyperparameter choices including training and testing data duration, network architecture, and algorithm choice affect model performance.more » « less
-
Surface meltwater generated on ice shelves fringing the Antarctic Ice Sheet can drive ice-shelf collapse, leading to ice sheet mass loss and contributing to global sea level rise. A quantitative assessment of supraglacial lake evolution is required to understand the influence of Antarctic surface meltwater on ice-sheet and ice-shelf stability. Cloud computing platforms have made the required remote sensing analysis computationally trivial, yet a careful evaluation of image processing techniques for pan-Antarctic lake mapping has yet to be performed. This work paves the way for automating lake identification at a continental scale throughout the satellite observational record via a thorough methodological analysis. We deploy a suite of different trained supervised classifiers to map and quantify supraglacial lake areas from multispectral Landsat-8 scenes, using training data generated via manual interpretation of the results from k-means clustering. Best results are obtained using training datasets that comprise spectrally diverse unsupervised clusters from multiple regions and that include rock and cloud shadow classes. We successfully apply our trained supervised classifiers across two ice shelves with different supraglacial lake characteristics above a threshold sun elevation of 20°, achieving classification accuracies of over 90% when compared to manually generated validation datasets. The application of our trained classifiers produces a seasonal pattern of lake evolution. Cloud shadowed areas hinder large-scale application of our classifiers, as in previous work. Our results show that caution is required before deploying ‘off the shelf’ algorithms for lake mapping in Antarctica, and suggest that careful scrutiny of training data and desired output classes is essential for accurate results. Our supervised classification technique provides an alternative and independent method of lake identification to inform the development of a continent-wide supraglacial lake mapping product.more » « less
-
Abstract Many studies of Earth surface processes and landscape evolution rely on having accurate and extensive data sets of surficial geologic units and landforms. Automated extraction of geomorphic features using deep learning provides an objective way to consistently map landforms over large spatial extents. However, there is no consensus on the optimal input feature space for such analyses. We explore the impact of input feature space for extracting geomorphic features from land surface parameters (LSPs) derived from digital terrain models (DTMs) using convolutional neural network (CNN)‐based semantic segmentation deep learning. We compare four input feature space configurations: (a) a three‐layer composite consisting of a topographic position index (TPI) calculated using a 50 m radius circular window, square root of topographic slope, and TPI calculated using an annulus with a 2 m inner radius and 10 m outer radius, (b) a single illuminating position hillshade, (c) a multidirectional hillshade, and (d) a slopeshade. We test each feature space input using three deep learning algorithms and four use cases: two with natural features and two with anthropogenic features. The three‐layer composite generally provided lower overall losses for the training samples, a higher F1‐score for the withheld validation data, and better performance for generalizing to withheld testing data from a new geographic extent. Results suggest that CNN‐based deep learning for mapping geomorphic features or landforms from LSPs is sensitive to input feature space. Given the large number of LSPs that can be derived from DTM data and the variety of geomorphic mapping tasks that can be undertaken using CNN‐based methods, we argue that additional research focused on feature space considerations is needed and suggest future research directions. We also suggest that the three‐layer composite implemented here can offer better performance in comparison to using hillshades or other common terrain visualization surfaces and is, thus, worth considering for different mapping and feature extraction tasks.more » « less
-
The burgeoning field of brain health research increasingly leverages artificial intelligence (AI) to analyze and interpret neuroimaging data. Medical foundation models have shown promise of superior performance with better sample efficiency. This work introduces a novel approach towards creating 3-dimensional (3D) medical foundation models for multimodal neuroimage segmentation through self-supervised training. Our approach involves a novel two-stage pretraining approach using vision transformers. The first stage encodes anatomical structures in generally healthy brains from the large-scale unlabeled neuroimage dataset of multimodal brain magnetic resonance imaging (MRI) images from 41,400 participants. This stage of pertaining focuses on identifying key features such as shapes and sizes of different brain structures. The second pretraining stage identifies disease-specific attributes, such as geometric shapes of tumors and lesions and spatial placements within the brain. This dual-phase methodology significantly reduces the extensive data requirements usually necessary for AI model training in neuroimage segmentation with the flexibility to adapt to various imaging modalities. We rigorously evaluate our model, BrainSegFounder, using the Brain Tumor Segmentation (BraTS) challenge and Anatomical Tracings of Lesions After Stroke v2.0 (ATLAS v2.0) datasets. BrainSegFounder demonstrates a significant performance gain, surpassing the achievements of the previous winning solutions using fully supervised learning. Our findings underscore the impact of scaling up both the model complexity and the volume of unlabeled training data derived from generally healthy brains. Both of these factors enhance the accuracy and predictive capabilities of the model in neuroimage segmentation tasks. Our pretrained models and code are at https://github.com/lab-smile/BrainSegFounder.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1117/12.2293968
