Medical Image Segmentation Techniques

This paper develops automated 3D segmentation techniques for human eye and orbit structures using MRI and evaluates axial length estimation.

This paper validates MRI brain volume measurements using ex vivo specimens and finds that SynthSeg provides accurate T1-weighted volumes compared to water displacement.

MortX provides a benchmark dataset for mouse cortex segmentation and registration, enabling domain generalization across postnatal ages.

SPROUT provides a user-friendly segmentation framework for volumetric images, achieving high-quality results with less manual input than supervised models.

The paper introduces Baby Open Brains (BOBs), an open-source repository of manually curated infant brain segmentations to improve reproducibility in neuroimaging research.

This study compares 2D and 3D U-Net for segmenting human aorta from CT images, finding 3D U-Net performs better but has geometry reconstruction errors.

LightMed improves medical image segmentation by using FFT-based processing to enhance robustness against noise and adversarial attacks.

The paper develops a deep learning method for mouse MRI brain segmentation using global positional encoding, achieving high accuracy and outperforming existing methods.

This paper introduces an automated workflow using the SAM model to segment cysts in histological images of ADPKD, showing increased diseased tissue proportion over time and outperforming existing tools.

AmygdalaGo-BOLT improves amygdala segmentation accuracy and efficiency in pediatric MRI scans across multiple centers and ages.

This paper introduces a deep learning model called Nailnet for segmenting fingernail plate and lunula from hand images, achieving high IoU and accuracy scores.

This paper introduces a confidence-enhanced semi-supervised learning framework for mediastinal neoplasm segmentation, improving performance by leveraging unlabeled data and refining prediction reliability.

This paper introduces a 3D morphological simulation approach to test phenotype detection methods, showing that detectability depends on effect size, sample size, and ROI selection.

This paper evaluates automated methods for hippocampal segmentation and HVR calculation, finding that a CNN-based method outperforms others in accuracy and reliability.

SAMson is an automated tool for rodent brain extraction from MRI data, outperforming existing methods in segmentation accuracy across diverse datasets.

The paper introduces NextBrain, a probabilistic histological atlas of the human brain, and presents a tool for automated MRI segmentation of anatomical regions.

UniFed addresses harmonisation and privacy in neuroimaging by enabling federated learning with partially labelled distributed data, producing high-quality segmentations.

The paper introduces a multi-contrast neonatal brain template and atlases to improve image processing and spatial normalization in neonatal neuroimaging.

This paper introduces a semi-supervised segmentation model for identifying supragranular and infragranular layers in human cerebral cortex MRI scans with high precision.

StARQ automates high-granularity segmentation and quantification of brainstem nuclei using deep learning, enhancing reliable computational analysis.

This paper proposes a sequential modelling approach for medical image segmentation using a modified MegaByte model, achieving comparable performance on intracranial haemorrhage segmentation and outperforming existing methods on tiny structures.

This paper develops a deep learning model using QSM to differentiate gold fiducial markers from calcification in prostate cancer patients, achieving higher accuracy than conventional MRI methods.

This study compares image registration techniques for functional ultrasound imaging, finding that Imregdeform provides the best alignment of 2D fUSI brain data.

The paper introduces a public dataset of neonatal brain MRIs with expert annotations for HIE-related lesions to improve segmentation methods for small, diffuse lesions.

AutoMacq is an automatic pipeline for processing rhesus macaque MRI data, producing reliable voxel-based and surface-based metrics with minimal manual intervention.

This paper introduces a topology-aware focal loss to improve 3D image segmentation by addressing topological errors through a Wasserstein distance-based constraint.

The paper introduces a deep learning pipeline for automated fetal brain segmentation and volumetry, showing reduced need for manual refinement and agreement with manual segmentations in clinical cases.

This paper introduces an axis-referenced morphometric model (ARMM) to reshape the hippocampus based on its anatomical structure, showing improved shape similarity and ability to capture neurodegenerative changes.

The paper presents a two-step methodology to reduce bias in structural MRI analysis of pediatric traumatic brain injury cases with lesions using Freesurfer, retaining 8 out of 14 cases.

The paper develops a deep learning model to automate fetal intracranial volume segmentation from 3D ultrasound and finds sex differences in ICV and growth.