Journal of biomedical physics & engineering · 2026
Detection of Valve Vegetations in Native and Prosthetic Valves using Echocardiographic Radiomics and Deep Learning on Transesophageal Echocardiography Images.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Mortazavi Seyed Mohammad Javad, Kadivar Fatemeh, Zare Shahrokh, Kanani Abolfazl, Mosleh-Shirazi Mohammad Amin, Masoudi Sadegh, Karimi-Busheri Feridoun, Mehrabani Davood
Аннотация
BACKGROUND: Infective Endocarditis (IE) is a life-threatening condition that requires rapid and accurate diagnosis. Transesophageal Echocardiography (TEE) is the gold standard for detecting valve vegetations; however, its interpretation is highly operator-dependent and particularly challenging in patients with prosthetic valves. Recent advances in artificial intelligence, especially deep learning, offer opportunities to improve diagnostic accuracy and reduce observer variability.
OBJECTIVE: This study aimed to evaluate the performance of deep learning-based models for detecting vegetations in TEE images to support the diagnostic workflow of IE.
MATERIAL AND METHODS: In this retrospective experimental study, a Faster Region-based Convolutional Neural Network (Faster R-CNN) was implemented to localize valve vegetations in TEE images. Four model configurations were developed using DenseNet121 and ResNet50 backbones, each trained in frozen and fine-tuned modes. All models were pretrained on RadImageNet. The dataset consisted of 1,000 annotated TEE frames acquired from both native and prosthetic heart valves.
RESULTS: The fine-tuned DenseNet121 model achieved the best performance, with a mean Average Precision (mAP) of 0.653 and an Area Under the Curve (AUC) of 0.858. Its frozen version demonstrated lower performance (mAP=0.416, AUC=0.640). The fine-tuned ResNet50 model reached a mAP of 0.593 and an AUC of 0.789, while the frozen ResNet50 showed the lowest performance (mAP=0.403, AUC=0.601).
CONCLUSION: Both fine-tuned DenseNet121 and ResNet50 models demonstrated effective localization of valve vegetations in TEE images, with comparable IoU performance. Although DenseNet121 showed superior classification accuracy, the similar localization results highlight the potential of both models as physician-assistive tools for enhancing IE diagnostic workflows.
OBJECTIVE: This study aimed to evaluate the performance of deep learning-based models for detecting vegetations in TEE images to support the diagnostic workflow of IE.
MATERIAL AND METHODS: In this retrospective experimental study, a Faster Region-based Convolutional Neural Network (Faster R-CNN) was implemented to localize valve vegetations in TEE images. Four model configurations were developed using DenseNet121 and ResNet50 backbones, each trained in frozen and fine-tuned modes. All models were pretrained on RadImageNet. The dataset consisted of 1,000 annotated TEE frames acquired from both native and prosthetic heart valves.
RESULTS: The fine-tuned DenseNet121 model achieved the best performance, with a mean Average Precision (mAP) of 0.653 and an Area Under the Curve (AUC) of 0.858. Its frozen version demonstrated lower performance (mAP=0.416, AUC=0.640). The fine-tuned ResNet50 model reached a mAP of 0.593 and an AUC of 0.789, while the frozen ResNet50 showed the lowest performance (mAP=0.403, AUC=0.601).
CONCLUSION: Both fine-tuned DenseNet121 and ResNet50 models demonstrated effective localization of valve vegetations in TEE images, with comparable IoU performance. Although DenseNet121 showed superior classification accuracy, the similar localization results highlight the potential of both models as physician-assistive tools for enhancing IE diagnostic workflows.