Coronary vessel segmentation in coronary angiography with a multi-scale U-shaped transformer incorporating boundary aggregation and topology preservation

Physics in Medicine and Biology, Год журнала: 2024, Номер 69(2), С. 025012 - 025012

Опубликована: Янв. 10, 2024

Abstract Coronary vessel segmentation plays a pivotal role in automating the auxiliary diagnosis of coronary heart disease. The continuity and boundary accuracy segmented vessels directly affect subsequent processing. Notably, during segmentation, with severe stenosis can easily cause errors breakage, resulting isolated islands. To address these issues, we propose novel multi-scale U-shaped transformer aggregation topology preservation (UT-BTNet) for angiography. Specifically, considering characteristics vessels, first develop UT-BTNet which combines advantages convolutional neural networks (CNN) transformer, is able to effectively extract local global features angiographic images. Secondly, innovatively employ loss topological two stages, addition traditional losses. In stage, adopted, has effect aggregation. second applied preserve after network converges. experiment, metrics Dice intersection over union (IoU), specifically (BIoU) Betti error evaluate results. results show that 0.9291, IoU 0.8687, BIoU 0.5094, 0.3400. Compared other state-of-the-art methods, achieves better results, while ensuring indicating its potential clinical value.

Язык: Английский

---

Few-Shot Learning for Medical Image Segmentation Using 3D U-Net and Model-Agnostic Meta-Learning (MAML)

Diagnostics, Год журнала: 2024, Номер 14(12), С. 1213 - 1213

Опубликована: Июнь 7, 2024

Deep learning has attained state-of-the-art results in general image segmentation problems; however, it requires a substantial number of annotated images to achieve the desired outcomes. In medical field, availability is often limited. To address this challenge, few-shot techniques have been successfully adapted rapidly generalize new tasks with only few samples, leveraging prior knowledge. paper, we employ gradient-based method known as Model-Agnostic Meta-Learning (MAML) for segmentation. MAML meta-learning algorithm that quickly adapts by updating model’s parameters based on limited set training samples. Additionally, use an enhanced 3D U-Net foundational network our models. The convolutional neural specifically designed We evaluate approach TotalSegmentator dataset, considering four tasks: liver, spleen, right kidney, and left kidney. demonstrate facilitates rapid adaptation using images. 10-shot settings, achieved mean dice coefficients 93.70%, 85.98%, 81.20%, 89.58% kidney segmentation, respectively. five-shot sittings, Dice 90.27%, 83.89%, 77.53%, 87.01% Finally, assess effectiveness proposed dataset collected from local hospital. Employing 90.62%, 79.86%, 79.87%, 78.21%

Язык: Английский

---

Advancements in Deep Learning for Medical Image Analysis: A Comprehensive Exploration of Techniques, Applications, and Future Prospects

Опубликована: Янв. 2, 2025

This chapter analyzes the evolving landscape of medical image analysis, with a particular emphasis on research-driven incorporation deep learning models. The paradigm change caused by these models is investigated in consideration its applicability disease detection, diagnosis, and treatment planning. focuses crucial function segmentation detecting characterizing anomalies across various imaging modalities. From clinical significance to precision medicine, impacts precise are special role specific plans actions because problems that exist this field, such as limited data availability computation limits, proposes collaborative techniques address them. aims solve present constraints envisioning future defined advanced augmentation, domain adaptation, multi-modal fusion, paving way for more robust widely applicable analysis. overall goal encourage responsible development implementation, which will lead improvements patient outcomes healthcare diagnostics

Язык: Английский

---

An active learning model based on image similarity for skin lesion segmentation

Neurocomputing, Год журнала: 2025, Номер 630, С. 129690 - 129690

Опубликована: Фев. 20, 2025

Язык: Английский

---

Deep learning for pancreas segmentation on computed tomography: a systematic review

Artificial Intelligence Review, Год журнала: 2025, Номер 58(8)

Опубликована: Май 3, 2025

Язык: Английский

---

AnatSwin: An anatomical structure-aware transformer network for cardiac MRI segmentation utilizing label images

Neurocomputing, Год журнала: 2024, Номер 577, С. 127379 - 127379

Опубликована: Фев. 6, 2024

Despite the extensive utilization of deep learning in medical image segmentation, achieved accuracy remains inadequate for clinical requirements due to scarcity annotated data, which constrains acquisition anatomical knowledge. Leveraging information is particularly advantageous especially multi-modal and cross-domain tasks. To better capture represent structures, we propose a Swin Transformer-based structure-aware network, AnatSwin, adopts unique approach by utilizing label images as inputs. Compared with gray-scale images, devoid intensity information, explicitly enhance representation shape spatial tissue relationships, offering valuable resources structures effectively allowing model concentrate on understanding morphological relationship cues. AnatSwin follows an encoder–decoder architecture, where encoder incorporates two branches that share weights. The Swin-Transformer block serves basic unit encoder, accepting both template (representing correct structure) pseudo (generated registration model) In order facilitate efficient interaction among features at same hierarchy, attention-based feature (FI) introduced. FI enhances model's ability structure promoting interactions within branches. Furthermore, decoder employs blocks learn relationships between ultimately improving segmentation performance. Experimental evaluations demonstrate proposed outperforms state-of-the-art models, highlighting its significant potential well optimizing tasks related segmentation. This work signifies promising step forward addressing challenges paves way further advancements field.

Язык: Английский

---

AOA-guided hyperparameter refinement for precise medical image segmentation

Alexandria Engineering Journal, Год журнала: 2025, Номер 120, С. 547 - 560

Опубликована: Фев. 24, 2025

Язык: Английский

---

A unified approach to medical image segmentation by leveraging mixed supervision and self and transfer learning (MIST)

Computerized Medical Imaging and Graphics, Год журнала: 2025, Номер 122, С. 102517 - 102517

Опубликована: Март 5, 2025

Язык: Английский

---

Event-based optical flow: Method categorisation and review of techniques that leverage deep learning

Neurocomputing, Год журнала: 2025, Номер unknown, С. 129899 - 129899

Опубликована: Март 1, 2025

Язык: Английский

---

Boundary aware microscopic hyperspectral pathology image segmentation network guided by information entropy weight

Frontiers in Oncology, Год журнала: 2025, Номер 15

Опубликована: Март 27, 2025

Introduction Accurate segmentation of lesion tissues in medical microscopic hyperspectral pathological images is crucial for enhancing early tumor diagnosis and improving patient prognosis. However, the complex structure indistinct boundaries present significant challenges achieving precise segmentation. Methods To address these challenges, we propose a novel method named BE-Net. It employs multi-scale strategy edge operators to capture fine details, while incorporating information entropy construct attention mechanisms that further strengthen representation relevant features. Specifically, first Laplacian Gaussian operator convolution boundary feature extraction block, which encodes gradient through improved detection emphasizes channel weights based on weighting. We designed grouped module optimize fusion process between encoder decoder, with goal details emphasizing representations. Finally, spatial block guide model most important locations regions. Result evaluate BE-Net image datasets gastric intraepithelial neoplasia mucosal intestinal metaplasia. Experimental results demonstrate outperforms other state-of-the-art methods terms accuracy preservation. Discussion This advance has implications field MHSIs Our code freely available at https://github.com/sharycao/BE-NET .

Язык: Английский

---