SPEMix: a lightweight method via superclass pseudo-label and efficient mixup for echocardiogram view classification

Frontiers in Artificial Intelligence, Год журнала: 2025, Номер 7

Опубликована: Янв. 8, 2025

In clinical, the echocardiogram is most widely used for diagnosing heart diseases. Different diseases are diagnosed based on different views of images, so efficient view classification can help cardiologists diagnose disease rapidly. Echocardiogram mainly divided into supervised and semi-supervised methods. The methods have worse generalization performance due to difficulty labeling echocardiographic while achieve acceptable results via a little labeled data. However, current faces challenges declining accuracy out-of-distribution data constrained by complex model structures in clinical application. To deal with above challenges, we proposed novel open-set method classification, SPEMix, which improve leveraging unlabeled Our SPEMix consists two core blocks, DAMix Block SP Block. generate mixed mask that focuses valuable regions echocardiograms at pixel level high-quality augmented data, improving accuracy. superclass pseudo-label from perspective probability distribution, pseudolabel. We also evaluate our Unity dataset CAMUS dataset. lightweight trained best publicly available TMED2 For first time, applied solve limits application architecture more efficiently.

Язык: Английский

---

ProtoASNet: Comprehensive evaluation and enhanced performance with uncertainty estimation for Aortic stenosis classification in echocardiography

Medical Image Analysis, Год журнала: 2025, Номер 103, С. 103600 - 103600

Опубликована: Апрель 24, 2025

Язык: Английский

---

RAMAS-Net: a module-optimized convolutional network model for aortic valve stenosis recognition in echocardiography

Frontiers in Medicine, Год журнала: 2025, Номер 12

Опубликована: Апрель 28, 2025

Aortic stenosis (AS) is a valvular heart disease that obstructs normal blood flow from the left ventricle to aorta due pathological changes in valve, leading impaired cardiac function. Echocardiography key diagnostic tool for AS; however, its accuracy influenced by inter-observer variability, operator experience, and image quality, which can result misdiagnosis. Therefore, alternative methods are needed assist healthcare professionals achieving more accurate diagnoses. We proposed deep learning model, RSMAS-Net, automated identification diagnosis of AS using echocardiography. The model enhanced ResNet50 backbone replacing Stage 4 with Spatial Channel Reconstruction Convolution (SCConv) Multi-Dconv Head Transposed Attention (MDTA) modules, aiming reduce redundant computations improve feature extraction capabilities. method was evaluated on TMED-2 echocardiography dataset, an 94.67%, F 1-score 94.37%, AUC 0.95 identification. Additionally, achieved 0.93 severity classification TMED-2. RSMAS-Net outperformed multiple baseline models recall, precision, parameter efficiency, inference time. It also 0.91 TMED-1 dataset. effectively diagnoses classifies echocardiographic images. integration SCConv MDTA modules enhances while reducing complexity compared original architecture. These results highlight potential improving assessment supporting clinical decision-making.

Язык: Английский

---

Rsmas-Net: A Deep Learning Model for Identifying Aortic Valve Stenosis in Static Echocardiography

Опубликована: Янв. 1, 2024

Язык: Английский

---

Application of transformers in stomatological imaging: A review

Digital Medicine, Год журнала: 2024, Номер 10(3)

Опубликована: Сен. 1, 2024

Stomatology extensively utilizes detailed imaging to assist clinicians, traditionally requiring labor-intensive manual analysis, which significantly adds their workload. Transformers are revolutionary neural network in deep learning, gaining substantial momentum computer vision tasks. Their introduction into medical imaging, particularly processing large image datasets, has been marked by remarkable precision and efficiency, establishing them as a pivotal tool emerging research. However, the application of stomatological is still its infancy. Current studies primarily focus on segmenting specific anatomical features such teeth jawbones, with some clinical implementations. Yet, comprehensive analytical potential this field remains largely untapped. This paper presents an introductory examination Transformers, coupled initial synthesis assessment dental applications across various areas. It will highlight observed advantages limitations contexts conclude discussion future research directions. serves foundational guide for in-depth investigations area.

Язык: Английский

---

Bioimage Informatics

Elsevier eBooks, Год журнала: 2024, Номер unknown

Опубликована: Янв. 1, 2024

Язык: Английский

---