Medical Image Segmentation Techniques

This paper constructs internally-consistent, fully-unbiased multimodal MRI brain templates from UK Biobank data, showing improved registration accuracy compared to existing templates.

The paper evaluates transformer models for binary semantic segmentation in CT scans, finding a redesigned Unet outperforms existing models with high precision and recall.

MEBRAINS 1.0 creates a multi-subject, multi-modal macaque brain atlas using T1 and T2 images, surface models, and curated segmentations to improve data integration.

MMORF is a nonlinear image registration tool for MRI brain images that outperforms existing methods in accuracy and distortion control across multiple domains.

This paper evaluates optimization and fine-tuning techniques for UNet variants in breast ultrasound segmentation, introducing Sharp Attention UNet which achieves high performance metrics.

The paper develops open-source tools for 3D analysis of brain dissection photographs and surface scans, achieving accurate reconstructions and segmentations comparable to MRI.

This study evaluates unsupervised clustering methods for segmenting dynamic total-body PET images at the organ level, finding that k-means with PCA performs best but still has limited accuracy.

The paper introduces a deep learning tool, SHIVA-WMH, for detecting white matter hyperintensities across different WMH severities and demonstrates its superior performance compared to existing methods.

This paper introduces an AI-based pipeline for analyzing cerebral organoids using MRI, achieving high segmentation accuracy and quality classification.

DeepSlice automatically registers mouse brain histology to the Allen Brain Atlas, achieving high accuracy and speed improvements of over 1000 times.

This paper develops an automated deep learning method for ischemic stroke lesion segmentation in mouse MRI, achieving high agreement with manual segmentation.

This paper develops a deep learning model to accurately segment the nucleus basalis of Meynert on 3T MRI, showing improved performance over standard atlases in both healthy subjects and TLE patients.

The paper releases curated anatomical fiducial placements for MRI datasets to improve image registration quality control and registration accuracy.

This paper introduces a surface-based hippocampal registration method that improves subfield alignment across different morphologies.

The paper introduces HumanBrainAtlas, a high-resolution in vivo MRI dataset with detailed segmentations for research, clinical, and educational use.

This paper develops an automated brain extraction method for piglet MRI images using Mask R-CNN, achieving high accuracy and consistency.

This paper introduces a Bayesian segmentation algorithm for thalamic nuclei using diffusion MRI and an improved histological atlas, showing improved accuracy in identifying nuclei and detecting Alzheimer's disease involvement.

The paper assesses the numerical and spatial reliability of hippocampal subfields and amygdala nuclei segmentation in FreeSurfer 7 using repeated scans from 923 participants.

This paper introduces MEMOS, a deep learning tool for segmenting 50 anatomical structures in mouse embryos, validated against atlas-based methods and knockout strain data.

This paper creates a digital 3D arterial territory atlas from stroke patient MRIs to aid in automated analysis of brain damage.

The paper compares manual and automated methods for hippocampus segmentation in infants, finding that ASHS provides more reliable results than manual or other automated approaches.

This paper presents a method to register light sheet microscopy data with magnetic resonance histology to restore in-skull morphology, achieving accurate registration with low displacement error.

This paper introduces month-specific brain atlases for infants aged two weeks to two years, integrating surface and volumetric data to capture early brain development in a unified coordinate system.

The paper introduces BEN, a deep learning method for brain tissue extraction across species and MRI modalities, showing improved accuracy and generalizability over existing tools.

HippUnfold is an automated BIDS-App for hippocampal unfolding and subfield segmentation using deep learning and topological constraints, enabling individual-specific morphometric analysis.

This paper introduces 3D-MASNet, a 3D mixed-scale asymmetric convolutional network for segmenting infant brain MR images, achieving high Dice coefficients in GM, WM, and CSF.

The paper introduces an automated pipeline for registering stand-alone histology sections to whole-brain MRI data with sub-voxel accuracy.

DeepParcellation provides accurate 3D MRI parcellation for older East Asians, outperforming existing methods in this demographic.

The paper proposes using the second derivative (DoG) for edge detection in MRI coregistration, improving alignment quality by generating closed surfaces and reducing intensity dependency.

The paper updates DeepImageTranslator with a tool for multimodal medical image segmentation analysis, showing it produces accurate statistics comparable to manual ROI selection.