On the usage of average Hausdorff distance for segmentation performance assessment: hidden error when used for ranking

European Radiology Experimental, Год журнала: 2021, Номер 5(1)

Опубликована: Янв. 21, 2021

Abstract Average Hausdorff distance is a widely used performance measure to calculate the between two point sets. In medical image segmentation, it compare ground truth images with segmentations allowing their ranking. We identified, however, ranking errors of average making less suitable for applications in segmentation assessment. To mitigate this error, we present modified calculation that have coined “balanced distance”. simulate ranking, manually created non-overlapping common magnetic resonance angiography cerebral vessel as our use-case. Adding consecutively and randomly truth, sets simulated increasing number errors. Each set was ranked using both measures. calculated Kendall rank correlation coefficient each segmentation. The rankings produced by balanced had significantly higher median (1.00) than those (0.89). 200 total rankings, former misranked 52 whilst latter 179 segmentations. Balanced more quality assessment distance.

Язык: Английский

---

BRAVE-NET: Fully Automated Arterial Brain Vessel Segmentation in Patients With Cerebrovascular Disease

Frontiers in Artificial Intelligence, Год журнала: 2020, Номер 3

Опубликована: Сен. 25, 2020

Introduction: Arterial brain vessel assessment is crucial for the diagnostic process in patients with cerebrovascular disease. Non-invasive neuroimaging techniques, such as time-of-flight (TOF) magnetic resonance angiography (MRA) imaging are applied clinical routine to depict arteries. They are, however, only visually assessed. Fully automated segmentation integrated into could facilitate time-critical diagnosis of abnormalities and might identification valuable biomarkers events. In present work, we developed validated a new deep learning model segmentation, coined BRAVE-NET, on large aggregated dataset diseases. Methods: BRAVE-NET multiscale 3-D convolutional neural network (CNN) 264 from three different studies enrolling A context path, dually capturing high- low-resolution volumes, supervision were implemented. The was compared baseline Unet variants paths supervision, respectively. models using high-quality manual labels ground truth. Next precision recall, performance assessed quantitatively by Dice coefficient (DSC); average Hausdorff distance (AVD); 95-percentile (95HD); via visual qualitative rating. Results: surpassed other arterial DSC = 0.931, AVD 0.165, 95HD 29.153. also most resistant toward false labelings revealed analysis. improvement primarily attributed integration multiscaling path lesser extent architectural component. Discussion: We state-of-the-art tailored pathology. provide an extensive experimental validation encompassing variability disease external set healthy volunteers. framework provides technological foundation improving workflow can serve biomarker extraction tool

Язык: Английский

---

UNet Deep Learning Architecture for Segmentation of Vascular and Non-Vascular Images: A Microscopic Look at UNet Components Buffered With Pruning, Explainable Artificial Intelligence, and Bias

IEEE Access, Год журнала: 2022, Номер 11, С. 595 - 645

Опубликована: Дек. 26, 2022

Biomedical image segmentation (BIS) task is challenging due to the variations in organ types, position, shape, size, scale, orientation, and contrast. Conventional methods lack accurate automated designs. Artificial intelligence (AI)-based UNet has recently dominated BIS. This first review of its kind that microscopically addressed types by complexity, stratification components, addressing vascular vs. non-vascular framework, key challenge UNet-based architecture, finally interfacing three facets AI, pruning, explainable AI (XAI), AI-bias. PRISMA was used select 267 studies. Five classes were identified labeled as conventional UNet, superior attention-channel hybrid ensemble UNet. We discovered 81 considering six kinds namely encoder, decoder, skip connection, bridge network, loss function, their combination. Vascular architecture compared. AP(ai)Bias 2.0-UNet these based on (i) attributes performance, (ii) and, (iii) pruning (compression). bias such ranking, radial, regional area, (iv) PROBAST, (v) ROBINS-I applied compared using a Venn diagram. systems with sUNet attention. Most studies suffered from low interest XAI strategies. None models qualified be bias-free. There need move paper-to-practice paradigms for clinical evaluation settings.

Язык: Английский

---

Imaging of the pial arterial vasculature of the human brain in vivo using high-resolution 7T time-of-flight angiography

eLife, Год журнала: 2022, Номер 11

Опубликована: Апрель 29, 2022

The pial arterial vasculature of the human brain is only blood supply to neocortex, but quantitative data on morphology and topology these mesoscopic arteries (diameter 50–300 µm) remains scarce. Because it commonly assumed that flow velocities in vessels are prohibitively slow, non-invasive time-of-flight magnetic resonance angiography (TOF-MRA)—which well suited high 3D imaging resolutions—has not been applied arteries. Here, we provide a theoretical framework outlines how TOF-MRA can visualize small vivo, by employing extremely voxels at size individual vessels. We then evidence for this theory 140 µm isotropic resolution using 7 Tesla (T) (MRI) scanner prospective motion correction, show one voxel width diameter be detected. conclude limited slow flow, instead achievable image resolution. This study represents first targeted, comprehensive account vivo brain. ultra-high-resolution will enable characterization vascular anatomy across investigate patterns relationships between functional architecture.

Язык: Английский

---

Comparing methods of detecting and segmenting unruptured intracranial aneurysms on TOF-MRAS: The ADAM challenge

NeuroImage, Год журнала: 2021, Номер 238, С. 118216 - 118216

Опубликована: Май 27, 2021

Accurate detection and quantification of unruptured intracranial aneurysms (UIAs) is important for rupture risk assessment to allow an informed treatment decision be made. Currently, 2D manual measures used assess UIAs on Time-of-Flight magnetic resonance angiographies (TOF-MRAs) lack 3D information there substantial inter-observer variability both aneurysm size growth. could helpful improve but are time-consuming would therefore benefit from a reliable automatic UIA segmentation method. The Aneurysm Detection segMentation (ADAM) challenge was organised in which methods were developed submitted evaluated diverse clinical TOF-MRA dataset. A training set (113 cases with total 129 UIAs) released, each case including TOF-MRA, structural MR image (T1, T2 or FLAIR), annotation any present UIA(s) the centre voxel UIA(s). test 141 (with 153 evaluation. Two tasks proposed: (1) (2) TOF-MRAs. Teams containerised set. Task 1 using metrics sensitivity false positive count. 2 dice similarity coefficient, modified hausdorff distance (95th percentile) volumetric similarity. For task, ranking made based average metrics. In total, eleven teams participated task nine those 2. won by method specifically designed (i.e. not participating 2). Based metrics, top two performed statistically significantly better than all other methods. performance top-ranking comparable visual inspection larger aneurysms. Segmentation method, after selection true UIAs, similar interobserver performance. ADAM remains open future submissions improved submissions, live leaderboard provide benchmarking developments at https://adam.isi.uu.nl/.

Язык: Английский

---

All answers are in the images: A review of deep learning for cerebrovascular segmentation

Computerized Medical Imaging and Graphics, Год журнала: 2023, Номер 107, С. 102229 - 102229

Опубликована: Апрель 7, 2023

Язык: Английский

---

3D Vascular Segmentation Supervised by 2D Annotation of Maximum Intensity Projection

IEEE Transactions on Medical Imaging, Год журнала: 2024, Номер 43(6), С. 2241 - 2253

Опубликована: Фев. 6, 2024

Vascular structure segmentation plays a crucial role in medical analysis and clinical applications. The practical adoption of fully supervised models is impeded by the intricacy time-consuming nature annotating vessels 3D space. This has spurred exploration weakly-supervised approaches that reduce reliance on expensive annotations. Despite this, existing weakly methods employed organ segmentation, which encompass points, bounding boxes, or graffiti, have exhibited suboptimal performance when handling sparse vascular structure. To alleviate this issue, we employ maximum intensity projection (MIP) to decrease dimensionality volume 2D image for efficient annotation, labels are utilized provide guidance oversight training vessel model. Initially, generate pseudo-labels blood using annotations projections. Subsequently, taking into account acquisition method labels, introduce network fuses 2D-3D deep features via MIP further improve performance. Furthermore, integrate confidence learning uncertainty estimation refine generated pseudo-labels, followed fine-tuning network. Our validated five datasets (including cerebral vessel, aorta coronary artery), demonstrating highly competitive segmenting potential significantly time effort required annotation. code available at: https://github.com/gzq17/Weakly-Supervised-by-MIP.

Язык: Английский

---

A scoping review of automatic and semi-automatic MRI segmentation in human brain imaging

Radiography, Год журнала: 2025, Номер 31(2), С. 102878 - 102878

Опубликована: Янв. 31, 2025

Язык: Английский

---

A dataset of rodent cerebrovasculature from in vivo multiphoton fluorescence microscopy imaging

Scientific Data, Год журнала: 2023, Номер 10(1)

Опубликована: Март 17, 2023

We present MiniVess, the first annotated dataset of rodent cerebrovasculature, acquired using two-photon fluorescence microscopy. MiniVess consists 70 3D image volumes with segmented ground truths. Segmentations were created traditional processing operations, a U-Net, and manual proofreading. Code for preprocessing steps U-Net are provided. Supervised machine learning methods have been widely used automated biomedical images. While much emphasis has placed on development new network architectures loss functions, there an increased need publicly available annotated, or segmented, datasets. Annotated datasets necessary during model training validation. In particular, that collected from different labs to test generalizability models. hope this will be helpful in testing reliability tools analyzing

Язык: Английский

---

Applicable artificial intelligence for brain disease: A survey

Neurocomputing, Год журнала: 2022, Номер 504, С. 223 - 239

Опубликована: Июль 16, 2022

Язык: Английский

---