International journal of computer assisted radiology and surgery · 2022
Automatic diagnosis of pneumothorax with M-mode ultrasound images based on D-MPL.
СтатьяОбзорНазвание и аннотация: Перевод пока не подготовлен
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.
Zhang Tao, Yan Shiju, Wei Gaofeng, Yang Linying, Yu Tianxiang, Ma Yibo
Аннотация
PURPOSE: To address the difficulties of M-mode ultrasound images classification in pneumothorax diagnosis and the shortcomings of existing neural network algorithms in this field, we proposed an M-mode ultrasound images classification model based on Disturbed Meta-Pseudo-Labels (D-MPL).
METHODS: An M-mode ultrasound image augmentation system was designed to make the model more robust and generalizable. In D-MPL, teacher-generated pseudo-labeling was first taught to students through a soft mask, and additional disturbance data were added to the teacher network. As the loss of the teacher network continues to decline, disturbance data were injected to improve the generalization of the model to cope with image differences across patients in clinical settings.
RESULTS: We compared the proposed model with four commonly used models, including MPL, EfficientnetB2, Inception V3, and Resnet101, in order to confirm its efficacy. Our model has an average specificity of 98.28%, sensitivity of 98.22%, F1-score of 98.23%, and AUC of 98.10%, according to the experiment findings, and its comprehensive performance is better than the above four models.
CONCLUSION: The results demonstrated our model's superiority over the competition and its greater. The model proposed in this study is expected to assist doctors in the diagnosis of pneumothorax as an auxiliary mean.
METHODS: An M-mode ultrasound image augmentation system was designed to make the model more robust and generalizable. In D-MPL, teacher-generated pseudo-labeling was first taught to students through a soft mask, and additional disturbance data were added to the teacher network. As the loss of the teacher network continues to decline, disturbance data were injected to improve the generalization of the model to cope with image differences across patients in clinical settings.
RESULTS: We compared the proposed model with four commonly used models, including MPL, EfficientnetB2, Inception V3, and Resnet101, in order to confirm its efficacy. Our model has an average specificity of 98.28%, sensitivity of 98.22%, F1-score of 98.23%, and AUC of 98.10%, according to the experiment findings, and its comprehensive performance is better than the above four models.
CONCLUSION: The results demonstrated our model's superiority over the competition and its greater. The model proposed in this study is expected to assist doctors in the diagnosis of pneumothorax as an auxiliary mean.