Weakly Supervised Intracranial Hemorrhage Segmentation using Head-Wise Gradient-Infused Self-Attention Maps from a Swin Transformer in Categorical Learning DOI Open Access
Amirhossein Rasoulian, Soorena Salari, Yiming Xiao

и другие.

The Journal of Machine Learning for Biomedical Imaging, Год журнала: 2023, Номер 2(MLCN 2022), С. 338 - 360

Опубликована: Авг. 29, 2023

Intracranial hemorrhage (ICH) is a life-threatening medical emergency that requires timely and accurate diagnosis for effective treatment improved patient survival rates. While deep learning techniques have emerged as the leading approach image analysis processing, most commonly employed supervised often large, high-quality annotated datasets can be costly to obtain, particularly pixel/voxel-wise segmentation. To address this challenge facilitate ICH decisions, we introduce novel weakly method segmentation, utilizing Swin transformer trained on an classification task with categorical labels. Our leverages hierarchical combination of head-wise gradient-infused self-attention maps generate Additionally, conducted exploratory study different strategies showed binary has more positive impact compared full subtyping. With mean Dice score 0.44, our technique achieved similar segmentation performance popular U-Net Swin-UNETR models supervision outperformed using GradCAM, demonstrating excellent potential proposed framework in challenging tasks. code available at <a href='https://github.com/HealthX-Lab/HGI-SAM'>https://github.com/HealthX-Lab/HGI-SAM</a>

Язык: Английский

Fast and Low-GPU-memory abdomen CT organ segmentation: The FLARE challenge DOI
Jun Ma, Yao Zhang, Song Gu

и другие.

Medical Image Analysis, Год журнала: 2022, Номер 82, С. 102616 - 102616

Опубликована: Сен. 13, 2022

Язык: Английский

Процитировано

102
Detection of cerebral aneurysms using artificial intelligence: a systematic review and meta-analysis DOI Creative Commons

Munaib Din,

Siddharth Agarwal,

Mariusz Grzeda

и другие.

Journal of NeuroInterventional Surgery, Год журнала: 2022, Номер 15(3), С. 262 - 271

Опубликована: Ноя. 14, 2022

Background Subarachnoid hemorrhage from cerebral aneurysm rupture is a major cause of morbidity and mortality. Early identification, aided by automated systems, may improve patient outcomes. Therefore, systematic review meta-analysis the diagnostic accuracy artificial intelligence (AI) algorithms in detecting aneurysms using CT, MRI or DSA was performed. Methods MEDLINE, Embase, Cochrane Library Web Science were searched until August 2021. Eligibility criteria included studies fully to detect MRI, CT DSA. Following Preferred Reporting Items for Systematic Reviews Meta-Analysis: Diagnostic Test Accuracy (PRISMA-DTA), articles assessed Quality Assessment Studies 2 (QUADAS-2). Meta-analysis bivariate random-effect model determine pooled sensitivity, specificity, area under receiver operator characteristic curve (ROC-AUC). PROSPERO: CRD42021278454. Results 43 included, 41/43 (95%) retrospective. 34/43 (79%) used AI as standalone tool, while 9/43 (21%) assisting reader. 23/43 (53%) deep learning. Most had high bias risk applicability concerns, limiting conclusions. Six gave (pooled) 91.2% (95% CI 82.2% 95.8%) sensitivity; 16.5% 9.4% 27.1%) false-positive rate (1-specificity); 0.936 ROC-AUC. Five reader-assistive 90.3% 88.0% – 92.2%) 7.9% 3.5% 16.8%) rate; 0.910 Conclusion has potential support clinicians aneurysms. Interpretation limited due poor generalizability. Multicenter, prospective are required assess clinical practice.

Язык: Английский

Процитировано

44
Automation bias in AI-assisted detection of cerebral aneurysms on time-of-flight MR angiography DOI Creative Commons
Su Hwan Kim, Severin Schramm,

Evamaria Olga Riedel

и другие.

La radiologia medica, Год журнала: 2025, Номер unknown

Опубликована: Фев. 12, 2025

Язык: Английский

Процитировано

2
Towards Automated Brain Aneurysm Detection in TOF-MRA: Open Data, Weak Labels, and Anatomical Knowledge DOI Creative Commons
Tommaso Di Noto, Guillaume Marie, Sébastien Tourbier

и другие.

Neuroinformatics, Год журнала: 2022, Номер 21(1), С. 21 - 34

Опубликована: Авг. 18, 2022

Abstract Brain aneurysm detection in Time-Of-Flight Magnetic Resonance Angiography (TOF-MRA) has undergone drastic improvements with the advent of Deep Learning (DL). However, performances supervised DL models heavily rely on quantity labeled samples, which are extremely costly to obtain. Here, we present a model for that overcomes issue “weak” labels: oversized annotations considerably faster create. Our weak labels resulted be four times generate than their voxel-wise counterparts. In addition, our leverages prior anatomical knowledge by focusing only plausible locations occurrence. We first train and evaluate through cross-validation an in-house TOF-MRA dataset comprising 284 subjects (170 females / 127 healthy controls 157 patients 198 aneurysms). On this dataset, best achieved sensitivity 83%, False Positive (FP) rate 0.8 per patient. To assess generalizability, then participated challenge data (93 patients, 20 controls, 125 public challenge, was 68% (FP = 2.5), ranking 4th/18 open leaderboard. found no significant difference between risk-of-rupture groups ( p 0.75), 0.72), or sizes 0.15). Data, code weights released under permissive licenses. demonstrate can alleviate necessity prohibitively expensive annotations.

Язык: Английский

Процитировано

28
Geometric Deep Learning Using Vascular Surface Meshes for Modality-Independent Unruptured Intracranial Aneurysm Detection DOI
Kimberley M. Timmins, Irene C. van der Schaaf, Iris N. Vos

и другие.

IEEE Transactions on Medical Imaging, Год журнала: 2023, Номер 42(11), С. 3451 - 3460

Опубликована: Июнь 23, 2023

Early detection of unruptured intracranial aneurysms (UIAs) enables better rupture risk and preventative treatment assessment. UIAs are usually diagnosed on Time-of-Flight Magnetic Resonance Angiographs (TOF-MRA) or contrast-enhanced Computed Tomography (CTA). Various automatic voxel-based deep learning UIA methods have been developed, but these limited to a single modality. We propose modality-independent method using geometric model with high resolution surface meshes brain vessels. A mesh convolutional neural network ResU-Net style architecture was used. performance investigated different input pooling resolutions, including additional edge features (shape index curvedness). Both higher (15,000 edges) curvature improved (average sensitivity: 65.6%, false positive count/image (FPC/image): 1.61). were detected in an independent TOF-MRA test set CTA average sensitivity 52.0% 48.3% FPC/image 1.04 1.05 respectively. provide deep-learning vascular comparable state-of-the-art methods.

Язык: Английский

Процитировано

15
Where is VALDO? VAscular Lesions Detection and segmentatiOn challenge at MICCAI 2021 DOI Creative Commons
Carole H. Sudre,

Kimberlin Van Wijnen,

Florian Dubost

и другие.

Medical Image Analysis, Год журнала: 2023, Номер 91, С. 103029 - 103029

Опубликована: Ноя. 19, 2023

Imaging markers of cerebral small vessel disease provide valuable information on brain health, but their manual assessment is time-consuming and hampered by substantial intra- interrater variability. Automated rating may benefit biomedical research, as well clinical assessment, diagnostic reliability existing algorithms unknown. Here, we present the results VAscular Lesions DetectiOn Segmentation (Where VALDO?) challenge that was run a satellite event at international conference Medical Image Computing Computer Aided Intervention (MICCAI) 2021. This aimed to promote development methods for automated detection segmentation sparse imaging disease, namely enlarged perivascular spaces (EPVS) (Task 1), microbleeds 2) lacunes presumed vascular origin 3) while leveraging weak noisy labels. Overall, 12 teams participated in proposing solutions one or more tasks (4 Task 1 - EPVS, 9 2 Microbleeds 6 3 Lacunes). Multi-cohort data used both training evaluation. Results showed large variability performance across tasks, with promising notably EPVS not practically useful yet Lacunes. It also highlighted inconsistency cases deter use an individual level, still proving population level.

Язык: Английский

Процитировано

14
Morphology-aware multi-source fusion–based intracranial aneurysms rupture prediction DOI
Chubin Ou, Caizi Li, Yi Qian

и другие.

European Radiology, Год журнала: 2022, Номер 32(8), С. 5633 - 5641

Опубликована: Фев. 18, 2022

Язык: Английский

Процитировано

21
StrokeNet: An automated approach for segmentation and rupture risk prediction of intracranial aneurysm DOI
Muhammad Irfan, Khalid Mahmood Malik, Jamil Ahmad

и другие.

Computerized Medical Imaging and Graphics, Год журнала: 2023, Номер 108, С. 102271 - 102271

Опубликована: Июль 22, 2023

Язык: Английский

Процитировано

12
Accelerated simulation methodologies for computational vascular flow modelling DOI Creative Commons
Michael MacRaild, Ali Sarrami‐Foroushani, Toni Lassila

и другие.

Journal of The Royal Society Interface, Год журнала: 2024, Номер 21(211)

Опубликована: Фев. 1, 2024

Vascular flow modelling can improve our understanding of vascular pathologies and aid in developing safe effective medical devices. models typically involve solving the nonlinear Navier-Stokes equations complex anatomies using physiological boundary conditions, often presenting a multi-physics multi-scale computational problem to be solved. This leads highly expensive that require excessive time. review explores accelerated simulation methodologies, specifically focusing on modelling. We reduced order (ROM) techniques like zero-/one-dimensional modal decomposition-based ROMs machine learning (ML) methods including ML-augmented ROMs, ML-based physics-informed ML models. discuss applicability each method acceleration effectiveness addressing domain-specific challenges. When available, we provide statistics accuracy speed-up factors for various applications related acceleration. Our findings indicate type model has strengths limitations depending context. To accelerate real-world problems, propose future research capable handling significant geometric variability inherent such problems.

Язык: Английский

Процитировано

5
Classification, detection, and segmentation performance of image-based AI in intracranial aneurysm: a systematic review DOI Creative Commons
Zhiyue Zhou,

Yuxuan Jin,

Haili Ye

и другие.

BMC Medical Imaging, Год журнала: 2024, Номер 24(1)

Опубликована: Июль 2, 2024

Abstract Background The detection and management of intracranial aneurysms (IAs) are vital to prevent life-threatening complications like subarachnoid hemorrhage (SAH). Artificial Intelligence (AI) can analyze medical images, CTA or MRA, spotting nuances possibly overlooked by humans. Early facilitates timely interventions improved outcomes. Moreover, AI algorithms offer quantitative data on aneurysm attributes, aiding in long-term monitoring assessing rupture risks. Methods We screened four databases (PubMed, Web Science, IEEE Scopus) for studies using artificial intelligence identify IA. Based algorithmic methodologies, we categorized them into classification, segmentation, combined, then their merits shortcomings compared. Subsequently, elucidate potential challenges that contemporary might encounter within real-world clinical diagnostic contexts. Then outline prospective research trajectories underscore key concerns this evolving field. Results Forty-seven IA recognition based were included search screening criteria. retrospective results represent current different modal images predict risk blockage. In diagnosis, effectively improve the accuracy reduce missed false positives. Conclusions algorithm detect unobtrusive more accurately communicating arteries cavernous sinus avoid further expansion. addition, analyzing blockage before after surgery help doctors plan treatment uncertainties process.

Язык: Английский

Процитировано

5