Medical Image Segmentation Techniques

This paper trains a Unet model to segment arteries and estimate diameters from T1w and T2w MRI images, showing T2w images can provide comparable results to TOF.

This study compares temporal ICA (tICA) and spatial ICA (sICA) for canine brain tumor segmentation, finding tICA achieves higher overlap with manual delineation than sICA.

The paper introduces funcmasker-flex, a BIDS App using a 3D U-net to automate fetal fMRI brain segmentation, achieving high Dice scores and reducing manual effort.

The paper introduces MINT, a tool for multimodal MRI registration, and presents an ABCD atlas for whole-brain analysis, showing it outperforms existing methods in speed while maintaining accuracy.

The paper introduces MiniVess, a rodent cerebrovasculature dataset for testing machine learning tools in biomedical image analysis.

FollicleFinder automates 3D segmentation of human ovarian follicles and measures morphological properties like diameter, surface area, and volume.

This paper develops a deep learning model for automated segmentation of the human Habenula from T1-weighted MRI, achieving high accuracy and robustness across different MRI protocols.

The paper creates 3D MRI atlases of normal and abnormal fetal aortic arch anatomy and evaluates deep learning for automated multi-label segmentation.

This paper presents a hybrid neuro-fuzzy system for automated gastric bleeding lesion segmentation in endoscopy images, achieving high accuracy and robustness against noise.

The paper assesses MRI-based volumetric changes in formalin-fixed Beagle brains over time and evaluates segmentation reproducibility across operators.

This paper benchmarks geometric and traditional deep learning models for cortical segmentation and neurodevelopmental phenotype prediction using neonatal cortical surface data.

This paper introduces a 3D U-Net method for automated skull stripping in mouse fMRI, achieving high accuracy and matching manual segmentation results.

This paper introduces a 3D U-Net framework for automatic brain extraction in rat MRI data, showing improved performance over existing methods.

DARQ is a deep learning method for automated quality control of stereotaxic MRI registration, achieving high accuracy comparable to manual assessment.

The paper develops a CNN-based method for automated hippocampus segmentation in rat MRI images post-TBI, showing superior performance compared to registration-based methods.

This study compares FSL-FIRST and FreeSurfer automated subcortical segmentation tools against manual segmentation in 5-year-olds, finding FSL-FIRST more accurate for most structures but with overestimation issues for small ones.

This paper presents a framework for deploying deep learning segmentation models for medical images across operating systems and languages, enabling clinical use and multi-institutional studies.

NiftyTorch provides a deep learning framework for neuroimaging tasks, aiming to simplify 3D deep learning for non-experts.

This paper applies unsupervised domain adaptation techniques to segment neonatal brain MRI data from a preterm population, improving segmentation accuracy and enabling analysis of cortical thickness and language outcomes.

This paper develops a method to reconstruct the cerebellar cortex from standard MRI using diffeomorphic registration of a high-resolution template, enabling accurate topographic reconstructions for neuroimaging applications.

The paper introduces a pipeline to automate manual mapping of cortical folds to human-interpretable shapes, demonstrating it on detecting discontinuities in the orbitofrontal cortex with high accuracy.

The paper develops a transfer-learning U-Net model for brain extraction in non-human primates using human data, achieving better performance than traditional methods and popular MRI packages.

The paper introduces PVSSAS, a tool for semi-automated segmentation of perivascular spaces, showing high agreement with manual methods.

This paper develops an automated segmentation pipeline for GM/WM in brain WSIs using CNNs and post-processing, achieving high IoU scores and reducing manual effort.

This study compares cortical and subcortical segmentation methods in Alzheimer's disease using MRI data, finding that HIPS, volBrain, and CAT perform well while BrainSuite shows misclassification issues.

This study compares Freesurfer and FSL automated MRI brain segmentation in a large longitudinal dataset, finding no linear correlation and differences in subcortical volume estimates.

The paper improves unsupervised tissue segmentation by integrating MSI and H&E image data, using morphological features and AMI score to evaluate segmentation validity.

The paper compares manual and automated ventricle segmentation in a rat model of schizophrenia to evaluate the effectiveness of automated methods for detecting structural changes.

The paper segments and analyzes the anterior Lamina Cribrosa surface from OCT images using a Bayesian method with non-local MRF and K-means, achieving 96.7% accuracy.

This paper introduces a transfer-learning approach using CNNs to segment infant brain structures, showing improved performance on newborns and 6-month-olds, especially in preterm infants.