Medical Image Segmentation Techniques

The paper compares automated lesion segmentation methods for chronic stroke T1-weighted MRI data, finding LINDA performs best overall while Gaussian Bayes has highest recall.

This paper introduces a Dilated Saliency U-Net for WMH segmentation using an irregularity age map, achieving better performance with fewer parameters and shorter training time.

This paper proposes a method for breast region segmentation in mammograms using multi-atlas deformable registration, achieving better performance than existing methods.

This paper introduces anatomical fiducials (AFIDs) to improve spatial correspondence in brain imaging and demonstrates their effectiveness over traditional voxel overlap metrics.

The paper evaluates enhanced machine learning techniques for segmenting ischaemic stroke lesions in MR perfusion images using a CNN, finding data augmentation with binary closing improves segmentation accuracy.

This paper proposes a brain tissue segmentation method using MP2RAGE multi-contrast images, showing faster processing and comparable segmentation accuracy to existing tools.

The paper updates the M-CRIB neonatal brain atlas to align with the Desikan-Killiany-Tourville adult atlas and FreeSurfer subcortical segmentations, resulting in a more compatible and accurate whole-brain parcellation.

The paper presents an automated pipeline for quantifying white matter hyperintensity in acute ischemic stroke cohorts, showing improved accuracy and enabling large-scale analysis.

This paper evaluates two automated methods for quantifying enlarged perivascular spaces in clinical MRI scans across multiple sites, finding one method achieves correlation with visual scoring comparable to inter-rater agreement in some regions.

This paper develops a variational framework for aligning histology stacks to an atlas, addressing curvature non-identifiability and enabling accurate reconstruction of whole-brain structures.

This paper introduces a method to map brain slices to a reference atlas without 3D reconstruction, achieving results comparable to experts and better than reconstruction-first approaches.

This paper introduces a diffeomorphic centroid method for shape analysis of large datasets, showing it achieves similar results to variational templates but with 70% faster computation.

This study evaluates two semi-automated hippocampal subfield segmentation methods across different age groups in a pediatric sample, finding negligible age-related differences in subfield validity.

This paper evaluates MS lesion segmentation algorithms using a new open-science infrastructure, finding that automatic methods lag behind human experts in detection and delineation.

The paper evaluates a statistical QA method for detecting cortical reconstruction errors in neuroimaging data and finds it less accurate than visual inspection but less resource-intensive.

The paper introduces a method to improve gray matter segmentation in ultra high field MRI by using 2D histograms, showing improved borders and providing an open implementation.

The paper constructs a high-resolution probabilistic in vivo atlas of human subcortical brain nuclei using MRI data from 168 adults and provides software tools for atlas development.

MIMoSA improves automated white matter lesion segmentation in multiple sclerosis by integrating multimodal MRI data through inter-modal coupling regression and local logistic regression.

The paper introduces ATLAS, a dataset of 304 T1-weighted MRIs with manual lesion segmentations to improve lesion segmentation algorithm accuracy.

The paper presents a 3D digital brain atlas of the common marmoset monkey with MRI and Nissl histology data, offering improved segmentation and file format for neuroscience research.

The paper introduces a fully automated pipeline for neonatal cortical surface reconstruction from MRI, achieving high accuracy in 85% of cases.

This study uses an Irregularity Age Map to adapt transfer learning for brain lesion segmentation and predict WMH progression/regression without manual labels.

This paper introduces a computational method to unfold the hippocampus, enabling 2D mapping of subfields and revealing structural regularities validated by histology.

This paper introduces a new regularization method for cortical surface alignment in brain imaging, improving the robustness and biological interpretability of multi-modal data alignment.

This study improves automated vein segmentation in MRI by combining susceptibility-weighted images, quantitative susceptibility maps, and a vein atlas into a composite vein image, showing significant performance gains.

The paper presents a high-resolution MRI template and analysis tools for macaque brains, enabling standardized processing and comparison across subjects and studies.

Mindcontrol is a web application that enables collaborative quality control of neuroimaging data through visualization, annotation, and integration with datasets.

Open source Mindboggle pipeline extracts cortical folds, sulci and fundi and computes diverse shape measures, yielding normative morphometry for 101 manually labeled healthy human brains.