Learning Pixel Level Affinity with Class Labels for Weakly Supervised Segmentation of Lung Cavities

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: April 7, 2025

Accurately annotating lung cavities (LCs) at the pixel level in computed tomography (CT) images presents a significant challenge due to their diverse shapes and sizes. To address this limitation, weakly supervised semantic segmentation (WSSS) methods utilizing sparse annotations, such as image-level labels, have emerged promising trend. This paper proposes novel scribble-supervised framework for LCs that leverages annotation-driven affinity. The introduces bidirectional interaction Mamba UNet model, named MambaUNeLCsT, designed inefficiency of transformer models processing long sequences. refine coarse pseudo-labels, an attention-based affinity pseudo-label refinement module is incorporated, employing algorithm establish associations between unlabeled pseudo-labeled samples. approach infers labels samples by computing sample similarities. Additionally, overcome limited spatial supervision provided scribble-based included, effectively capturing complete morphology boundary information LCs. enhances model’s capability recognize process fine structures. Experimental results demonstrate MambaUNeLCsT achieves state-of-the-art performance 3D medical image segmentation, outperforming existing WSSS tasks.

Language: Английский

---

SAM-LCA: a computationally efficient SAM-based model for tuberculosis detection in chest X-rays

Multimedia Systems, Journal Year: 2025, Volume and Issue: 31(3)

Published: April 26, 2025

Language: Английский

---

An extensive analysis of artificial intelligence and segmentation methods transforming cancer recognition in medical imaging

Biomedical Physics & Engineering Express, Journal Year: 2024, Volume and Issue: 10(4), P. 045046 - 045046

Published: June 7, 2024

Abstract Recent advancements in computational intelligence, deep learning, and computer-aided detection have had a significant impact on the field of medical imaging. The task image segmentation, which involves accurately interpreting identifying content an image, has garnered much attention. main objective this is to separate objects from background, thereby simplifying enhancing significance image. However, existing methods for segmentation their limitations when applied certain types images. This survey paper aims highlight importance techniques by providing thorough examination advantages disadvantages. accurate cancer regions images crucial ensuring effective treatment. In study, we also extensive analysis Computer-Aided Diagnosis (CAD) systems identification, with focus recent research advancements. critically assesses various compares effectiveness. Convolutional neural networks (CNNs) attracted particular interest due ability segment classify large datasets, thanks capacity self- learning decision-making.

Language: Английский

---

MCBERT: A Multi-Modal Framework for the Diagnosis of Autism Spectrum Disorder

Biological Psychology, Journal Year: 2024, Volume and Issue: unknown, P. 108976 - 108976

Published: Dec. 1, 2024

Language: Английский

---

SwinUNeCCt: Bidirectional Hash-based Agent Transformer for Cervical Cancer MRI Image Multi-task Learning

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Aug. 7, 2024

Background: Cervical cancer is the fourth most common malignant tumor among women globally, posing a significant threat to women's health. In 2022, approximately 600,000 new cases were reported, and 340,000 deaths occurred due cervical cancer. Magnetic resonance imaging (MRI) preferred method for diagnosing, staging, evaluating However, manual segmentation of MRI images time-consuming subjective. Therefore, there an urgent need automatic models identify lesions in scans accurately. Methods: All magnetic our research are from patients diagnosed by pathology at Tongren City People's Hospital. Strict data selection criteria clearly defined inclusion exclusion conditions established ensure consistency accuracy results. The dataset contains 122 patients, with each patient having 100 pelvic dynamic contrast-enhanced scans. Annotations jointly completed medical professionals Universiti Putra Malaysia Radiology Department Hospital reliability. Additionally, novel computer-aided diagnosis model named SwinUNeCCt proposed. This incorporates: i) A bidirectional hash-based agent multi-head self-attention mechanism, which optimizes interaction between local global features MRI, aiding more accurate lesion identification. ii) Reduced computational complexity mechanism. Results: effectiveness has been validated through comparisons state-of-the-art 3D models, including nnUnet, TransBTS, nnFormer, UnetR, UnesT, SwinUNetR, SwinUNeLCsT. semantic tasks without classification module, demonstrates excellent performance across multiple key metrics: achieving 95HD 6.25, IoU 0.669, DSC 0.802, all best results compared models. Simultaneously, strikes good balance efficiency complexity, requiring only 442.7 GFLOPs power 71.2M parameters. Furthermore, that include also exhibits powerful recognition capabilities. Although this slightly increases overhead its surpasses other comparative Conclusions: tasks, metrics. It balances well, maintaining high even module.

Language: Английский

---

Echocardiographic mitral valve segmentation model

Journal of King Saud University - Computer and Information Sciences, Journal Year: 2024, Volume and Issue: 36(9), P. 102218 - 102218

Published: Oct. 19, 2024

---