A transformation uncertainty and multi-scale contrastive learning-based semi-supervised segmentation method for oral cavity-derived cancer DOI Creative Commons
Ran Wang, Chengqi Lyu, Laihang Yu

и другие.

Frontiers in Oncology, Год журнала: 2025, Номер 15

Опубликована: Май 9, 2025

Objectives Oral cavity-derived cancer pathological images (OPI) are crucial for diagnosing oral squamous cell carcinoma (OSCC), but existing deep learning methods OPI segmentation rely heavily on large, accurately labeled datasets, which labor- and resource-intensive to obtain. This paper presents a semi-supervised method mitigate the limitations of scarce data by leveraging both unlabeled data. Materials We use Hematoxylin Eosin (H&E)-stained dataset (OCDC), consists 451 with tumor regions annotated verified pathologists. Our combines transformation uncertainty multi-scale contrastive learning. The estimation evaluates model’s confidence transformed via different methods, reducing discrepancies between teacher student models. Multi-scale enhances class similarity separability while teacher-student model similarity, encouraging diverse feature representations. Additionally, boundary-aware enhanced U-Net is proposed capture boundary information improve accuracy. Results Experimental results OCDC demonstrate that our outperforms fully supervised approaches, achieving superior performance. Conclusions method, integrating uncertainty, learning, U-Net, effectively addresses scarcity improves approach reduces dependency large promoting application AI in OSCC detection improving efficiency accuracy clinical diagnoses OSCC.

Язык: Английский

Semi-supervised Strong-Teacher Consistency Learning for few-shot cardiac MRI image segmentation DOI

Yuting Qiu,

James Meng,

Baihua Li

и другие.

Computer Methods and Programs in Biomedicine, Год журнала: 2025, Номер unknown, С. 108613 - 108613

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

1
Correlation-based switching mean teacher for semi-supervised medical image segmentation DOI
Guocheng DENG, Hao Sun, Wei Xie

и другие.

Neurocomputing, Год журнала: 2025, Номер unknown, С. 129818 - 129818

Опубликована: Март 1, 2025

Язык: Английский

Процитировано

1
Semantic Image Segmentation Employing U-Net-Based Ensemble Model DOI

M. Murali

Advances in computational intelligence and robotics book series, Год журнала: 2025, Номер unknown, С. 305 - 328

Опубликована: Март 7, 2025

Image segmentation is an important topic in computer vision, playing role wide range of applications such as medical image analysis, scene understanding, tumour boundary extraction among many others. aims to identify groups pixels and parts images that are similar belong together Semantic a classification with labels partitioning the objects. By applying segmentation, we can all objects image. The brain dataset utilizing for BRATS'20, which contains 317 images. proposed ensemble approach combining U-Net variants Mask RCNN models outperforms individual models. While method yielded improved Dice scores, using union six other methods achieved highest accuracy, indicated by superior scores. Specifically, model score 71.10 IoU 81.98. Additionally, demonstrated strong performance terms precision, reaching 84.96, recall value 81.90.

Язык: Английский

Процитировано

0
Dual prototypes contrastive learning based semi-supervised segmentation method for intelligent medical applications DOI
Tao Yue, Rongtao Xu,

Jingqian Wu

и другие.

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 154, С. 110905 - 110905

Опубликована: Май 1, 2025

Язык: Английский

Процитировано

0
A transformation uncertainty and multi-scale contrastive learning-based semi-supervised segmentation method for oral cavity-derived cancer DOI Creative Commons
Ran Wang, Chengqi Lyu, Laihang Yu

и другие.

Frontiers in Oncology, Год журнала: 2025, Номер 15

Опубликована: Май 9, 2025

Objectives Oral cavity-derived cancer pathological images (OPI) are crucial for diagnosing oral squamous cell carcinoma (OSCC), but existing deep learning methods OPI segmentation rely heavily on large, accurately labeled datasets, which labor- and resource-intensive to obtain. This paper presents a semi-supervised method mitigate the limitations of scarce data by leveraging both unlabeled data. Materials We use Hematoxylin Eosin (H&E)-stained dataset (OCDC), consists 451 with tumor regions annotated verified pathologists. Our combines transformation uncertainty multi-scale contrastive learning. The estimation evaluates model’s confidence transformed via different methods, reducing discrepancies between teacher student models. Multi-scale enhances class similarity separability while teacher-student model similarity, encouraging diverse feature representations. Additionally, boundary-aware enhanced U-Net is proposed capture boundary information improve accuracy. Results Experimental results OCDC demonstrate that our outperforms fully supervised approaches, achieving superior performance. Conclusions method, integrating uncertainty, learning, U-Net, effectively addresses scarcity improves approach reduces dependency large promoting application AI in OSCC detection improving efficiency accuracy clinical diagnoses OSCC.

Язык: Английский

Процитировано

0